{"id":98388,"date":"2026-10-12T12:55:45","date_gmt":"2026-10-12T09:55:45","guid":{"rendered":"https:\/\/twodots.gr\/?p=98388"},"modified":"2026-10-12T12:56:52","modified_gmt":"2026-10-12T09:56:52","slug":"ibm-granite-42-reasoning-ai-agents-enterprise","status":"publish","type":"post","link":"https:\/\/twodots.gr\/de\/ibm-granite-42-reasoning-ai-agents-enterprise\/","title":{"rendered":"IBM Granite 4.2: reasoning \u03ba\u03b1\u03b9 AI agents \u03bc\u03b5 \u03b1\u03bd\u03bf\u03b9\u03c7\u03c4\u03ac enterprise \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1"},"content":{"rendered":"<div class=\"td-article-lede\">\n<p><strong>\u0391\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03c0\u03c1\u03ce\u03c4\u03b1:<\/strong> \u03a4\u03bf IBM Granite 4.2 \u03c3\u03c5\u03bd\u03b4\u03c5\u03ac\u03b6\u03b5\u03b9 reasoning, native tool calling \u03ba\u03b1\u03b9 \u03b1\u03bd\u03bf\u03b9\u03c7\u03c4\u03ac \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 3B, 8B \u03ba\u03b1\u03b9 30B \u03b3\u03b9\u03b1 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf\u03c5\u03c2 enterprise AI agents \u03c3\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac workflows. \u0397 \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03ae \u03b1\u03be\u03af\u03b1 \u03c4\u03bf\u03c5 \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03cc\u03c4\u03b9 \u00ab\u03c3\u03ba\u03ad\u03c6\u03c4\u03b5\u03c4\u03b1\u03b9\u00bb \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03bf \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 \u03b1\u03af\u03c4\u03b7\u03bc\u03b1, \u03b1\u03bb\u03bb\u03ac \u03cc\u03c4\u03b9 \u03b5\u03c0\u03b9\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03c3\u03b5 \u03bc\u03b9\u03b1 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03bd\u03b1 \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03b9 \u03c0\u03bb\u03ae\u03c1\u03b5\u03c2 thinking, low-effort reasoning \u03ae \u03ac\u03bc\u03b5\u03c3\u03b7 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03c3\u03c5\u03bd\u03b4\u03ad\u03b5\u03b9 \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bc\u03b5 \u03b1\u03c5\u03c3\u03c4\u03b7\u03c1\u03ac \u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03b1 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1.<\/p>\n<p>\u0393\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7, \u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03b4\u03b5\u03bd \u03ba\u03c1\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03bc\u03cc\u03bd\u03bf \u03b1\u03c0\u03cc \u03c4\u03bf\u03bd \u03b1\u03c1\u03b9\u03b8\u03bc\u03cc \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03c4\u03c1\u03c9\u03bd. \u03a7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7 task success, latency, \u03ba\u03cc\u03c3\u03c4\u03bf\u03c5\u03c2 \u03c5\u03c0\u03bf\u03b4\u03bf\u03bc\u03ae\u03c2, \u03b1\u03ba\u03c1\u03af\u03b2\u03b5\u03b9\u03b1\u03c2 tool calls, \u03c3\u03c5\u03bc\u03c0\u03b5\u03c1\u03b9\u03c6\u03bf\u03c1\u03ac\u03c2 \u03c3\u03c4\u03b1 \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03ac \u03ba\u03b1\u03b9 \u03c3\u03c5\u03c7\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7\u03c2 \u03c0\u03b1\u03c1\u03ad\u03bc\u03b2\u03b1\u03c3\u03b7\u03c2. \u03a4\u03b1 8B \u03ba\u03b1\u03b9 30B \u03ad\u03c7\u03bf\u03c5\u03bd \u03c0\u03b5\u03c1\u03ac\u03c3\u03b5\u03b9 \u03b5\u03be\u03b5\u03b9\u03b4\u03b9\u03ba\u03b5\u03c5\u03bc\u03ad\u03bd\u03bf agentic reinforcement learning, \u03b5\u03bd\u03ce \u03c4\u03bf 3B \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03b5\u03b9 \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf \u03c9\u03c2 \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf\u03c2 \u03c5\u03c0\u03bf\u03c8\u03ae\u03c6\u03b9\u03bf\u03c2 \u03b3\u03b9\u03b1 \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03b1 workloads.<\/p>\n<\/div>\n<div class=\"td-article-toc\">\n<div class=\"td-toc-title\">\u03a0\u03b5\u03c1\u03b9\u03b5\u03c7\u03cc\u03bc\u03b5\u03bd\u03b1<\/div>\n<ul>\n<li><a href=\"#reasoning-prin-apo-tin-praxi\">Reasoning \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03c0\u03c1\u03ac\u03be\u03b7<\/a><\/li>\n<li><a href=\"#tria-megethi\">\u03a4\u03c1\u03af\u03b1 \u03bc\u03b5\u03b3\u03ad\u03b8\u03b7, \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03bf\u03af \u03c1\u03cc\u03bb\u03bf\u03b9<\/a><\/li>\n<li><a href=\"#architektoniki-context\">Dense \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae \u03ba\u03b1\u03b9 context<\/a><\/li>\n<li><a href=\"#pretraining-sft\">\u0391\u03c0\u03cc \u03c4\u03b1 15 \u03c4\u03c1\u03b9\u03c3. tokens \u03c3\u03c4\u03bf SFT<\/a><\/li>\n<li><a href=\"#agentic-rl\">Agentic RL \u03c3\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd\u03c4\u03b1<\/a><\/li>\n<li><a href=\"#tool-calling\">Native tool calling \u03bc\u03b5 \u03cc\u03c1\u03b9\u03b1<\/a><\/li>\n<li><a href=\"#benchmarks\">\u03a4\u03b9 \u03b4\u03b5\u03af\u03c7\u03bd\u03bf\u03c5\u03bd \u03c4\u03b1 benchmarks<\/a><\/li>\n<li><a href=\"#deployment-quantization\">Deployment \u03ba\u03b1\u03b9 quantization<\/a><\/li>\n<li><a href=\"#enterprise-governance\">\u03a4\u03b9 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ad\u03bd\u03b1 enterprise \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1<\/a><\/li>\n<li><a href=\"#pilot\">\u0395\u03c0\u03c4\u03ac \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03b3\u03b9\u03b1 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 pilot<\/a><\/li>\n<li><a href=\"#symperasma\">\u0391\u03bd\u03bf\u03b9\u03c7\u03c4\u03cc \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf, \u03cc\u03c7\u03b9 \u03ad\u03c4\u03bf\u03b9\u03bc\u03b7 \u03b1\u03c5\u03c4\u03bf\u03bd\u03bf\u03bc\u03af\u03b1<\/a><\/li>\n<\/ul>\n<\/div>\n<h2 id=\"reasoning-prin-apo-tin-praxi\">Reasoning \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03c0\u03c1\u03ac\u03be\u03b7<\/h2>\n<p>\u0397 IBM \u03c0\u03b1\u03c1\u03bf\u03c5\u03c3\u03b9\u03ac\u03b6\u03b5\u03b9 \u03c4\u03b1 Granite 4.2 \u03c9\u03c2 reasoning-focused \u03bf\u03b9\u03ba\u03bf\u03b3\u03ad\u03bd\u03b5\u03b9\u03b1 \u03b3\u03b9\u03b1 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2 \u03cc\u03c0\u03bf\u03c5 \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af \u03bd\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03ba\u03b1\u03bb\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7. \u03a0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03ba\u03b1\u03c4\u03b1\u03bd\u03bf\u03b5\u03af \u03ad\u03bd\u03b1\u03bd \u03c3\u03c4\u03cc\u03c7\u03bf, \u03bd\u03b1 \u03b1\u03bd\u03b1\u03ba\u03c4\u03ac \u03c0\u03bb\u03b7\u03c1\u03bf\u03c6\u03bf\u03c1\u03af\u03b1, \u03bd\u03b1 \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03b9 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf, \u03bd\u03b1 \u03b5\u03ba\u03c4\u03b5\u03bb\u03b5\u03af \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03c3\u03c4\u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03c3\u03b5\u03b9\u03c1\u03ac \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03b9 \u03c4\u03bf \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1. \u0391\u03c5\u03c4\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03b7 \u03bf\u03c5\u03c3\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae \u03c3\u03cd\u03bd\u03b4\u03b5\u03c3\u03b7 \u03bc\u03b5 AI agents, coding assistants, search-driven flows \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03bf\u03cd\u03c2 \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2.<\/p>\n<p>\u039a\u03b1\u03b9 \u03c4\u03b1 \u03c4\u03c1\u03af\u03b1 \u03bc\u03b5\u03b3\u03ad\u03b8\u03b7 \u03c5\u03c0\u03bf\u03c3\u03c4\u03b7\u03c1\u03af\u03b6\u03bf\u03c5\u03bd thinking, non-thinking \u03ba\u03b1\u03b9 low-effort thinking \u03bc\u03ad\u03c3\u03c9 \u03c4\u03bf\u03c5 chat template. \u03a4\u03bf \u03c0\u03bb\u03ae\u03c1\u03b5\u03c2 thinking \u03c0\u03c1\u03bf\u03bf\u03c1\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b3\u03b9\u03b1 \u03c3\u03cd\u03bd\u03b8\u03b5\u03c4\u03b5\u03c2, \u03c0\u03bf\u03bb\u03c5\u03b2\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2. \u03a4\u03bf low-effort \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf reasoning budget \u03b3\u03b9\u03b1 \u03b5\u03c5\u03ba\u03bf\u03bb\u03cc\u03c4\u03b5\u03c1\u03b1 \u03b1\u03b9\u03c4\u03ae\u03bc\u03b1\u03c4\u03b1, \u03b5\u03bd\u03ce \u03c4\u03bf non-thinking \u03c0\u03b1\u03c1\u03b1\u03ba\u03ac\u03bc\u03c0\u03c4\u03b5\u03b9 \u03c4\u03b7\u03bd \u03b5\u03ba\u03c4\u03b5\u03c4\u03b1\u03bc\u03ad\u03bd\u03b7 \u03c3\u03c5\u03bb\u03bb\u03bf\u03b3\u03b9\u03c3\u03c4\u03b9\u03ba\u03ae. \u0397 \u03b4\u03c5\u03bd\u03b1\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b1\u03c5\u03c4\u03ae \u03b5\u03c0\u03b9\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 routing \u03b1\u03bd\u03ac \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b5\u03bd \u03b5\u03b3\u03b3\u03c5\u03ac\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03bc\u03cc\u03bd\u03b7 \u03c4\u03b7\u03c2 \u03c7\u03b1\u03bc\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03bf \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03ae \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1.<\/p>\n<p>\u0397 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03ae \u03b5\u03c6\u03b1\u03c1\u03bc\u03bf\u03b3\u03ae \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03bd\u03cc\u03bd\u03b1 \u03b4\u03c1\u03bf\u03bc\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7\u03c2 \u03c0\u03bf\u03c5 \u03bd\u03b1 \u03c3\u03c4\u03b7\u03c1\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c3\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1. \u039c\u03b9\u03b1 \u03c4\u03b1\u03be\u03b9\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7 \u03c0\u03c1\u03bf\u03ca\u03cc\u03bd\u03c4\u03bf\u03c2 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03bc\u03b7 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 thinking, \u03b5\u03bd\u03ce \u03b7 \u03b4\u03b9\u03b5\u03c1\u03b5\u03cd\u03bd\u03b7\u03c3\u03b7 \u03b5\u03bd\u03cc\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03bf\u03c5 order flow \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b1\u03c0\u03b1\u03b9\u03c4\u03b5\u03af \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03b1 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03ba\u03b1\u03b9 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1. \u0397 \u03b1\u03bd\u03ac\u03bb\u03c5\u03c3\u03b7 \u03b3\u03b9\u03b1 \u03c4\u03b1 <a href=\"https:\/\/twodots.gr\/ai-agent-harnesses-elegchos-kostos-self-hosting\/\">AI agent harnesses, \u03c4\u03bf self-hosting \u03ba\u03b1\u03b9 \u03c4\u03bf\u03bd \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf \u03ba\u03cc\u03c3\u03c4\u03bf\u03c5\u03c2<\/a> \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03b3\u03b9\u03b1\u03c4\u03af \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b5\u03af\u03bd\u03b1\u03b9 \u03bc\u03cc\u03bd\u03bf \u03ad\u03bd\u03b1 \u03bc\u03ad\u03c1\u03bf\u03c2 \u03c4\u03bf\u03c5 \u03c3\u03c5\u03c3\u03c4\u03ae\u03bc\u03b1\u03c4\u03bf\u03c2.<\/p>\n<h2 id=\"tria-megethi\">\u03a4\u03c1\u03af\u03b1 \u03bc\u03b5\u03b3\u03ad\u03b8\u03b7, \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03bf\u03af \u03c1\u03cc\u03bb\u03bf\u03b9<\/h2>\n<p>\u0397 \u03bf\u03b9\u03ba\u03bf\u03b3\u03ad\u03bd\u03b5\u03b9\u03b1 \u03ba\u03c5\u03ba\u03bb\u03bf\u03c6\u03bf\u03c1\u03b5\u03af \u03c3\u03b5 dense, decoder-only \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 3B, 8B \u03ba\u03b1\u03b9 30B \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03c4\u03c1\u03c9\u03bd \u03bc\u03b5 \u03ac\u03b4\u03b5\u03b9\u03b1 Apache 2.0. \u0397 \u03ba\u03bf\u03b9\u03bd\u03ae \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae \u03b4\u03b9\u03b5\u03c5\u03ba\u03bf\u03bb\u03cd\u03bd\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03b5\u03bd\u03b9\u03b1\u03af\u03b1 \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03af\u03b1 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7\u03c2, \u03b1\u03bb\u03bb\u03ac \u03c4\u03b1 training paths \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b9\u03c3\u03bf\u03b4\u03cd\u03bd\u03b1\u03bc\u03b1. \u03a4\u03b1 8B \u03ba\u03b1\u03b9 30B \u03c0\u03b5\u03c1\u03bd\u03bf\u03cd\u03bd \u03c4\u03bf \u03c0\u03bb\u03ae\u03c1\u03b5\u03c2 agentic RL block \u03b3\u03b9\u03b1 software engineering, terminal \u03ba\u03b1\u03b9 search, \u03b5\u03bd\u03ce \u03c4\u03bf 3B \u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03b5\u03af \u03c3\u03c5\u03bd\u03c4\u03bf\u03bc\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae foundational RL \u03ba\u03b1\u03b9 alignment.<\/p>\n<p>\u03a4\u03bf 3B \u03b5\u03af\u03bd\u03b1\u03b9 \u03bb\u03bf\u03b3\u03b9\u03ba\u03cc \u03bd\u03b1 \u03b4\u03bf\u03ba\u03b9\u03bc\u03b1\u03c3\u03c4\u03b5\u03af \u03c3\u03b5 extraction, \u03c4\u03b1\u03be\u03b9\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7, \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf edge inference \u03ba\u03b1\u03b9 \u03b1\u03c0\u03bb\u03ad\u03c2 \u03ba\u03bb\u03ae\u03c3\u03b5\u03b9\u03c2 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03c9\u03bd. \u03a4\u03bf 8B \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03b5\u03af \u03b9\u03c3\u03bf\u03c1\u03c1\u03bf\u03c0\u03b7\u03bc\u03ad\u03bd\u03bf \u03c3\u03b7\u03bc\u03b5\u03af\u03bf \u03b5\u03ba\u03ba\u03af\u03bd\u03b7\u03c3\u03b7\u03c2 \u03cc\u03c4\u03b1\u03bd \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 reasoning \u03bc\u03b5 \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b1\u03c0\u03b1\u03af\u03c4\u03b7\u03c3\u03b7 \u03c5\u03c0\u03bf\u03b4\u03bf\u03bc\u03ae\u03c2. \u03a4\u03bf 30B \u03c3\u03c4\u03bf\u03c7\u03b5\u03cd\u03b5\u03b9 \u03c3\u03b5 \u03b2\u03b1\u03b8\u03cd\u03c4\u03b5\u03c1\u03b5\u03c2 coding \u03ba\u03b1\u03b9 agentic \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2, \u03b1\u03bb\u03bb\u03ac \u03b7 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03ad\u03c7\u03b5\u03b9 \u03bd\u03cc\u03b7\u03bc\u03b1 \u03bc\u03cc\u03bd\u03bf \u03b1\u03bd \u03b7 \u03b1\u03cd\u03be\u03b7\u03c3\u03b7 \u03c4\u03bf\u03c5 task success \u03b4\u03b9\u03ba\u03b1\u03b9\u03bf\u03bb\u03bf\u03b3\u03b5\u03af latency, \u03bc\u03bd\u03ae\u03bc\u03b7 \u03ba\u03b1\u03b9 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b9\u03ba\u03cc \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2.<\/p>\n<div class=\"td-comparison td-comparison-cards--horizontal\">\n<div class=\"td-chart-head\">\n<p class=\"td-chart-kicker\">\u039c\u03af\u03b1 \u03bf\u03b9\u03ba\u03bf\u03b3\u03ad\u03bd\u03b5\u03b9\u03b1, \u03c4\u03c1\u03af\u03b1 deployment tiers<\/p>\n<p class=\"td-chart-title\">\u0397 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03b5\u03af \u03c4\u03b7\u03bd \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03bf \u03b1\u03c0\u03bf\u03b4\u03b5\u03ba\u03c4\u03cc \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2<\/p>\n<p class=\"td-chart-intro\">\u03a4\u03b1 3B, 8B \u03ba\u03b1\u03b9 30B \u03bc\u03bf\u03b9\u03c1\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c4\u03b7 \u03b2\u03b1\u03c3\u03b9\u03ba\u03ae \u03c3\u03c7\u03b5\u03b4\u03af\u03b1\u03c3\u03b7, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b5\u03bd \u03ad\u03c7\u03bf\u03c5\u03bd \u03af\u03b4\u03b9\u03bf training path, footprint \u03ae \u03ba\u03b1\u03c4\u03ac\u03bb\u03bb\u03b7\u03bb\u03bf \u03c1\u03cc\u03bb\u03bf.<\/p>\n<\/div>\n<div class=\"td-comparison-grid\">\n<div class=\"td-platform-card\">\n<p class=\"td-platform-name\">Granite 4.2 3B<\/p>\n<p>\u03a5\u03c0\u03bf\u03c8\u03ae\u03c6\u03b9\u03bf \u03b3\u03b9\u03b1 \u03c5\u03c8\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03bf throughput, \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03b1 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03ba\u03b1\u03b9 edge \u03ae \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03b1 on-premises deployments. \u0394\u03b5\u03bd \u03ad\u03bb\u03b1\u03b2\u03b5 \u03c4\u03bf \u03b5\u03be\u03b5\u03b9\u03b4\u03b9\u03ba\u03b5\u03c5\u03bc\u03ad\u03bd\u03bf agentic RL \u03c4\u03c9\u03bd \u03b4\u03cd\u03bf \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03c9\u03bd \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03c9\u03bd.<\/p>\n<\/div>\n<div class=\"td-platform-card\">\n<p class=\"td-platform-name\">Granite 4.2 8B<\/p>\n<p>\u039c\u03b5\u03c3\u03b1\u03af\u03b1 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03bc\u03b5 agentic RL, reasoning \u03ba\u03b1\u03b9 tool calling. \u039a\u03b1\u03c4\u03ac\u03bb\u03bb\u03b7\u03bb\u03bf \u03c0\u03c1\u03ce\u03c4\u03bf benchmark \u03b3\u03b9\u03b1 \u03bf\u03bc\u03ac\u03b4\u03b5\u03c2 \u03c0\u03bf\u03c5 \u03b8\u03ad\u03bb\u03bf\u03c5\u03bd \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf agent \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03be\u03b5\u03ba\u03b9\u03bd\u03bf\u03cd\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf footprint.<\/p>\n<\/div>\n<div class=\"td-platform-card\">\n<p class=\"td-platform-name\">Granite 4.2 30B<\/p>\n<p>\u0397 \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03c4\u03b7\u03c2 \u03c3\u03b5\u03b9\u03c1\u03ac\u03c2 \u03b3\u03b9\u03b1 \u03c3\u03cd\u03bd\u03b8\u03b5\u03c4\u03bf coding \u03ba\u03b1\u03b9 \u03b2\u03b1\u03b8\u03cd\u03c4\u03b5\u03c1\u03b1 multi-step tasks. \u0391\u03c0\u03b1\u03b9\u03c4\u03b5\u03af \u03b1\u03c5\u03c3\u03c4\u03b7\u03c1\u03cc capacity planning \u03ba\u03b1\u03b9 \u03b1\u03c0\u03cc\u03b4\u03b5\u03b9\u03be\u03b7 \u03cc\u03c4\u03b9 \u03c4\u03bf \u03c0\u03c1\u03cc\u03c3\u03b8\u03b5\u03c4\u03bf compute \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03bd\u03b5\u03b9 \u03c4\u03bf \u03b4\u03b9\u03ba\u03cc \u03c3\u03b1\u03c2 outcome.<\/p>\n<\/div>\n<\/div>\n<\/div>\n<p>\u0388\u03bd\u03b1 tiered \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c3\u03c4\u03ad\u03bb\u03bd\u03b5\u03b9 \u03c4\u03bf\u03bd \u03bc\u03b5\u03b3\u03ac\u03bb\u03bf \u03cc\u03b3\u03ba\u03bf \u03b1\u03c0\u03bb\u03ce\u03bd \u03b1\u03b9\u03c4\u03b7\u03bc\u03ac\u03c4\u03c9\u03bd \u03c3\u03c4\u03bf \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03ba\u03bb\u03b9\u03bc\u03b1\u03ba\u03ce\u03bd\u03b5\u03b9 \u03bc\u03cc\u03bd\u03bf \u03c4\u03b9\u03c2 \u03b1\u03b2\u03ad\u03b2\u03b1\u03b9\u03b5\u03c2 \u03ae \u03c5\u03c8\u03b7\u03bb\u03bf\u03cd \u03c1\u03af\u03c3\u03ba\u03bf\u03c5 \u03c0\u03b5\u03c1\u03b9\u03c0\u03c4\u03ce\u03c3\u03b5\u03b9\u03c2. \u03a0\u03c1\u03cc\u03ba\u03b5\u03b9\u03c4\u03b1\u03b9 \u03b3\u03b9\u03b1 \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03c0\u03bf\u03c5 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b4\u03bf\u03ba\u03b9\u03bc\u03b1\u03c3\u03c4\u03b5\u03af \u03c3\u03c4\u03bf \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc workload, \u03cc\u03c7\u03b9 \u03b3\u03b9\u03b1 \u03ba\u03b1\u03b8\u03bf\u03bb\u03b9\u03ba\u03ae \u03c3\u03cd\u03c3\u03c4\u03b1\u03c3\u03b7 \u03c4\u03b7\u03c2 IBM.<\/p>\n<h2 id=\"architektoniki-context\">Dense \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae \u03ba\u03b1\u03b9 context<\/h2>\n<p>\u03a4\u03b1 Granite 4.2 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd dense transformer \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae, Grouped Query Attention, RoPE \u03bc\u03b5 \u03b8 10.000.000, SwiGLU MLPs, RMSNorm \u03ba\u03b1\u03b9 \u03be\u03b5\u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03ac input\/output embeddings \u03c3\u03b5 bfloat16. \u03a4\u03bf 3B \u03ba\u03b1\u03b9 \u03c4\u03bf 8B \u03ad\u03c7\u03bf\u03c5\u03bd 40 layers, \u03b5\u03bd\u03ce \u03c4\u03bf 30B \u03ad\u03c7\u03b5\u03b9 64. \u03a3\u03b5 \u03b1\u03bd\u03c4\u03af\u03b8\u03b5\u03c3\u03b7 \u03bc\u03b5 mixture-of-experts \u03c3\u03c5\u03c3\u03c4\u03ae\u03bc\u03b1\u03c4\u03b1, \u03b5\u03bd\u03b5\u03c1\u03b3\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd \u03c4\u03bf dense parameter set \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 token, \u03ac\u03c1\u03b1 \u03b7 \u03bc\u03bd\u03ae\u03bc\u03b7, \u03c4\u03bf latency \u03ba\u03b1\u03b9 \u03c4\u03bf concurrency \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c1\u03b5\u03b1\u03bb\u03b9\u03c3\u03c4\u03b9\u03ba\u03cc load test.<\/p>\n<p>\u03a4\u03b1 \u03b5\u03c0\u03af\u03c3\u03b7\u03bc\u03b1 \u03c5\u03bb\u03b9\u03ba\u03ac \u03b8\u03ad\u03bb\u03bf\u03c5\u03bd \u03c0\u03c1\u03bf\u03c3\u03b5\u03ba\u03c4\u03b9\u03ba\u03ae \u03b1\u03bd\u03ac\u03b3\u03bd\u03c9\u03c3\u03b7 \u03c9\u03c2 \u03c0\u03c1\u03bf\u03c2 \u03c4\u03bf context. \u03a4\u03bf \u03c4\u03b5\u03c7\u03bd\u03b9\u03ba\u03cc \u03ac\u03c1\u03b8\u03c1\u03bf \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03b7 \u03c0\u03ad\u03bc\u03c0\u03c4\u03b7 \u03c6\u03ac\u03c3\u03b7 pre-training \u03c4\u03c9\u03bd Granite 4.1 base models \u03b5\u03c0\u03ad\u03ba\u03c4\u03b5\u03b9\u03bd\u03b5 \u03c4\u03bf context \u03c3\u03c4\u03b1 512K tokens, \u03ba\u03b1\u03b9 \u03c4\u03bf summary \u03c4\u03bf\u03c5 model card \u03bc\u03b9\u03bb\u03ac \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03b3\u03b9\u03b1 512K. \u038c\u03bc\u03c9\u03c2 \u03bf \u03c0\u03af\u03bd\u03b1\u03ba\u03b1\u03c2 \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae\u03c2 \u03ba\u03b1\u03b9 \u03c4\u03b1 \u03c0\u03b1\u03c1\u03b1\u03b4\u03b5\u03af\u03b3\u03bc\u03b1\u03c4\u03b1 serving \u03b3\u03b9\u03b1 vLLM \u03ba\u03b1\u03b9 SGLang \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd 131.072 tokens. \u0395\u03c0\u03bf\u03bc\u03ad\u03bd\u03c9\u03c2, \u03b7 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ae\u03c2 \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03cd\u03c3\u03b5\u03c4\u03b5 \u03c4\u03bf \u03c5\u03c0\u03bf\u03c3\u03c4\u03b7\u03c1\u03b9\u03b6\u03cc\u03bc\u03b5\u03bd\u03bf \u03cc\u03c1\u03b9\u03bf \u03c3\u03c4\u03bf \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf checkpoint, runtime \u03ba\u03b1\u03b9 quantization \u03b1\u03bd\u03c4\u03af \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03c6\u03ad\u03c1\u03b5\u03c4\u03b5 \u03ad\u03bd\u03b1\u03bd \u03b1\u03c1\u03b9\u03b8\u03bc\u03cc \u03b1\u03c0\u03cc \u03c4\u03bf marketing copy.<\/p>\n<aside class=\"td-article-note\"><strong>Context \u03b4\u03b5\u03bd \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 retrieval:<\/strong> \u03b1\u03ba\u03cc\u03bc\u03b7 \u03ba\u03b1\u03b9 \u03cc\u03c4\u03b1\u03bd \u03ad\u03bd\u03b1 runtime \u03b4\u03ad\u03c7\u03b5\u03c4\u03b1\u03b9 \u03c0\u03bf\u03bb\u03cd \u03bc\u03b5\u03b3\u03ac\u03bb\u03bf input, \u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03c3\u03b5 \u03bc\u03b1\u03ba\u03c1\u03cd \u03c3\u03c5\u03bc\u03b2\u03cc\u03bb\u03b1\u03b9\u03bf \u03ae knowledge base \u03b5\u03be\u03b1\u03c1\u03c4\u03ac\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc retrieval quality, \u03b8\u03ad\u03c3\u03b7 \u03c4\u03b7\u03c2 \u03c0\u03bb\u03b7\u03c1\u03bf\u03c6\u03bf\u03c1\u03af\u03b1\u03c2, distractors, prompt structure \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03b8\u03ad\u03c3\u03b9\u03bc\u03b7 \u03bc\u03bd\u03ae\u03bc\u03b7 KV.<\/aside>\n<p>\u0393\u03b9\u03b1 RAG \u03ba\u03b1\u03b9 long-document assistants, \u03bc\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 grounded accuracy \u03c3\u03b5 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ac \u03bc\u03ae\u03ba\u03b7 \u03ba\u03b1\u03b9 \u03b8\u03ad\u03c3\u03b5\u03b9\u03c2 evidence. \u0397 \u03c3\u03c7\u03b5\u03c4\u03b9\u03ba\u03ae \u03b1\u03bd\u03ac\u03bb\u03c5\u03c3\u03b7 \u03b3\u03b9\u03b1 \u03c4\u03b7 <a href=\"https:\/\/twodots.gr\/minima-kv-long-context-ai-cache-compression\/\">\u03c3\u03c5\u03bc\u03c0\u03af\u03b5\u03c3\u03b7 KV cache \u03c3\u03b5 long-context AI<\/a> \u03b5\u03be\u03b7\u03b3\u03b5\u03af \u03b3\u03b9\u03b1\u03c4\u03af \u03b7 \u03bf\u03bd\u03bf\u03bc\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae \u03c7\u03c9\u03c1\u03b7\u03c4\u03b9\u03ba\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b4\u03b5\u03bd \u03c4\u03b1\u03c5\u03c4\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03c4\u03bf \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03b5\u03be\u03c5\u03c0\u03b7\u03c1\u03ad\u03c4\u03b7\u03c3\u03b7\u03c2 \u03ae \u03c4\u03b7 \u03b4\u03b9\u03b1\u03c4\u03ae\u03c1\u03b7\u03c3\u03b7 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2.<\/p>\n<h2 id=\"pretraining-sft\">\u0391\u03c0\u03cc \u03c4\u03b1 15 \u03c4\u03c1\u03b9\u03c3. tokens \u03c3\u03c4\u03bf SFT<\/h2>\n<p>\u03a4\u03b1 Granite 4.2 \u03b5\u03af\u03bd\u03b1\u03b9 post-trained \u03c0\u03ac\u03bd\u03c9 \u03c3\u03c4\u03b1 Granite 4.1 base models. \u03a3\u03cd\u03bc\u03c6\u03c9\u03bd\u03b1 \u03bc\u03b5 \u03c4\u03bf \u03c4\u03b5\u03c7\u03bd\u03b9\u03ba\u03cc \u03ac\u03c1\u03b8\u03c1\u03bf \u03c4\u03b7\u03c2 IBM, \u03b7 \u03b2\u03ac\u03c3\u03b7 \u03b5\u03ba\u03c0\u03b1\u03b9\u03b4\u03b5\u03cd\u03c4\u03b7\u03ba\u03b5 \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03b1\u03c1\u03c7\u03ae \u03c3\u03b5 \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 15 \u03c4\u03c1\u03b9\u03c3\u03b5\u03ba\u03b1\u03c4\u03bf\u03bc\u03bc\u03cd\u03c1\u03b9\u03b1 tokens \u03bc\u03b5 \u03c0\u03ad\u03bd\u03c4\u03b5 \u03c6\u03ac\u03c3\u03b5\u03b9\u03c2: foundational pre-training, mid-training \u03bc\u03b5 \u03c3\u03c4\u03b1\u03b4\u03b9\u03b1\u03ba\u03ac \u03c0\u03b9\u03bf \u03b5\u03c0\u03b9\u03bc\u03b5\u03bb\u03b7\u03bc\u03ad\u03bd\u03b1 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03c6\u03ac\u03c3\u03b7 long-context. \u0397 IBM \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 \u03ad\u03bd\u03b1 \u03c4\u03c1\u03b9\u03c3\u03b5\u03ba\u03b1\u03c4\u03bf\u03bc\u03bc\u03cd\u03c1\u03b9\u03bf tokens \u03c3\u03c5\u03bd\u03b8\u03b5\u03c4\u03b9\u03ba\u03bf\u03cd \u03ba\u03ce\u03b4\u03b9\u03ba\u03b1 \u03b1\u03c0\u03cc \u03c4\u03bf CodeAlchemy \u03c9\u03c2 \u03bc\u03ad\u03c1\u03bf\u03c2 \u03c4\u03b7\u03c2 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b7\u03c2 \u03c3\u03b5 coding \u03ba\u03b1\u03b9 reasoning.<\/p>\n<p>\u03a4\u03bf supervised fine-tuning \u03c0\u03b5\u03c1\u03b9\u03bb\u03ac\u03bc\u03b2\u03b1\u03bd\u03b5 \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 7,2 \u03b5\u03ba\u03b1\u03c4\u03bf\u03bc\u03bc\u03cd\u03c1\u03b9\u03b1 \u03b4\u03b5\u03af\u03b3\u03bc\u03b1\u03c4\u03b1 \u03ae 100 \u03b4\u03b9\u03c3\u03b5\u03ba\u03b1\u03c4\u03bf\u03bc\u03bc\u03cd\u03c1\u03b9\u03b1 tokens, \u03b1\u03c0\u03cc \u03c4\u03b1 \u03bf\u03c0\u03bf\u03af\u03b1 \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 65 \u03b4\u03b9\u03c3\u03b5\u03ba\u03b1\u03c4\u03bf\u03bc\u03bc\u03cd\u03c1\u03b9\u03b1 \u03ae\u03c4\u03b1\u03bd trainable. \u03a4\u03bf mixture \u03ae\u03c4\u03b1\u03bd 31,6% agentic \u03ba\u03b1\u03b9 68,4% non-agentic. \u039c\u03ad\u03c3\u03b1 \u03c3\u03c4\u03bf agentic \u03c4\u03bc\u03ae\u03bc\u03b1, \u03c4\u03bf software engineering \u03b1\u03bd\u03c4\u03b9\u03c3\u03c4\u03bf\u03b9\u03c7\u03bf\u03cd\u03c3\u03b5 \u03c3\u03c4\u03bf 69%, \u03bc\u03b1\u03b6\u03af \u03bc\u03b5 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1 \u03b3\u03b9\u03b1 tool calling, terminal, \u03bc\u03b1\u03b8\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac, search \u03ba\u03b1\u03b9 action.<\/p>\n<p>\u039f\u03b9 trajectories \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03ae\u03b8\u03b7\u03ba\u03b1\u03bd \u03bc\u03b5 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ac harnesses, \u03cc\u03c0\u03c9\u03c2 OpenHands, OpenCode, Terminus-2, SWE-agent, MiniSWE, Codex \u03ba\u03b1\u03b9 Goose. \u0397 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b4\u03b5\u03bd \u03b1\u03c6\u03ad\u03b8\u03b7\u03ba\u03b5 \u03c3\u03b5 \u03ad\u03bd\u03b1 \u03b5\u03bd\u03b9\u03b1\u03af\u03bf \u03c6\u03af\u03bb\u03c4\u03c1\u03bf: \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03ae\u03b8\u03b7\u03ba\u03b1\u03bd LLM judges, \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf\u03b9 fabricated \u03c0\u03bb\u03b7\u03c1\u03bf\u03c6\u03bf\u03c1\u03b9\u03ce\u03bd \u03ba\u03b1\u03b9 invalid tool calls \u03ba\u03b1\u03b9 SHA-256 deduplication \u03c3\u03c4\u03b1 \u03c0\u03b5\u03b4\u03af\u03b1 tools \u03ba\u03b1\u03b9 messages. \u03a4\u03bf \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03cc \u03bc\u03ac\u03b8\u03b7\u03bc\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03cc\u03c4\u03b9 \u03b7 agentic \u03b9\u03ba\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03c9\u03bd, \u03c0\u03b5\u03c1\u03b9\u03b2\u03b1\u03bb\u03bb\u03cc\u03bd\u03c4\u03c9\u03bd, verifiers \u03ba\u03b1\u03b9 feedback loops, \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf\u03c5 parameter count.<\/p>\n<h2 id=\"agentic-rl\">Agentic RL \u03c3\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd\u03c4\u03b1<\/h2>\n<p>\u039c\u03b5\u03c4\u03ac \u03c4\u03bf SFT, \u03b7 IBM \u03b5\u03c6\u03ac\u03c1\u03bc\u03bf\u03c3\u03b5 \u03b4\u03b9\u03b1\u03b4\u03bf\u03c7\u03b9\u03ba\u03ac RL runs \u03c0\u03bf\u03c5 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03bf\u03cd\u03bd \u03c9\u03c2 warm-starts: RLVR, skill boosters, software engineering, terminal, search \u03ba\u03b1\u03b9 \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc RLHF. \u03a4\u03b1 8B \u03ba\u03b1\u03b9 30B \u03c0\u03b5\u03c1\u03bd\u03bf\u03cd\u03bd \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03bf \u03c4\u03bf agentic block. \u0397 \u03b5\u03ba\u03c0\u03b1\u03af\u03b4\u03b5\u03c5\u03c3\u03b7 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af asynchronous GRPO, \u03bc\u03b5 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03bf\u03cd\u03c2 workers \u03b3\u03b9\u03b1 generation \u03ba\u03b1\u03b9 policy updates \u03ce\u03c3\u03c4\u03b5 \u03b7 \u03c5\u03c0\u03bf\u03b4\u03bf\u03bc\u03ae \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae\u03c2 trajectories \u03bd\u03b1 \u03bc\u03b7 \u03bc\u03ad\u03bd\u03b5\u03b9 \u03b1\u03b4\u03c1\u03b1\u03bd\u03ae\u03c2 \u03ba\u03b1\u03c4\u03ac \u03c4\u03b1 optimizer steps.<\/p>\n<p>\u03a3\u03c4\u03bf software engineering stage, \u03ba\u03ac\u03b8\u03b5 task \u03c4\u03bf\u03c0\u03bf\u03b8\u03b5\u03c4\u03b5\u03af\u03c4\u03b1\u03b9 \u03c3\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc repository \u03ba\u03b1\u03b9 sandbox. \u03a4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b4\u03b9\u03b1\u03b2\u03ac\u03b6\u03b5\u03b9 \u03ba\u03ce\u03b4\u03b9\u03ba\u03b1, \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03b1\u03c1\u03c7\u03b5\u03af\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03c1\u03ad\u03c7\u03b5\u03b9 tests \u03bc\u03ad\u03c3\u03c9 OpenHands, \u03b5\u03bd\u03ce \u03c4\u03bf reward \u03c3\u03c5\u03bd\u03b4\u03ad\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 hidden tests. \u03a3\u03c4\u03bf terminal stage \u03c7\u03b5\u03b9\u03c1\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 live shell \u03bc\u03ad\u03c3\u03c9 Harbor\/Terminus-2, \u03c0\u03b1\u03c1\u03b1\u03c4\u03b7\u03c1\u03b5\u03af \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03ba\u03b1\u03b9 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03ae\u03c3\u03b5\u03b9 recovery \u03b3\u03b9\u03b1 \u03ad\u03c9\u03c2 64 environment turns. \u03a3\u03c4\u03bf search stage \u03b5\u03ba\u03c4\u03b5\u03bb\u03b5\u03af live web searches \u03b3\u03b9\u03b1 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03b5\u03c2 multi-hop \u03b5\u03c1\u03c9\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 \u03b7 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03b5\u03af\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc judge.<\/p>\n<p>\u0391\u03c5\u03c4\u03cc\u03c2 \u03bf \u03c3\u03c7\u03b5\u03b4\u03b9\u03b1\u03c3\u03bc\u03cc\u03c2 \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03b5\u03b9 \u03b1\u03c5\u03c4\u03cc\u03bc\u03b1\u03c4\u03b7 \u03b1\u03c3\u03c6\u03ac\u03bb\u03b5\u03b9\u03b1 \u03c3\u03b5 production. \u0394\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03cc\u03bc\u03c9\u03c2 \u03b3\u03b9\u03b1\u03c4\u03af \u03ad\u03bd\u03b1 \u03c3\u03bf\u03b2\u03b1\u03c1\u03cc enterprise test \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac tools, permissions, timeouts, \u03bb\u03b1\u03bd\u03b8\u03b1\u03c3\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03c9\u03bd \u03ba\u03b1\u03b9 \u03ba\u03c1\u03b9\u03c4\u03ae\u03c1\u03b9\u03b1 \u03c4\u03b5\u03c1\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd. \u0397 <a href=\"https:\/\/twodots.gr\/agentroom-ai-coding-agents-syntoni-parallilismos\/\">\u03c3\u03c5\u03bd\u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 AI coding agents<\/a> \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 coordination \u03ba\u03b1\u03b9 \u03ba\u03bf\u03b9\u03bd\u03ae \u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7, \u03b5\u03bd\u03ce \u03b7 \u03b5\u03ba\u03c0\u03b1\u03af\u03b4\u03b5\u03c5\u03c3\u03b7 \u03b5\u03bd\u03cc\u03c2 \u03bc\u03b5\u03bc\u03bf\u03bd\u03c9\u03bc\u03ad\u03bd\u03bf\u03c5 model \u03b4\u03b5\u03bd \u03bb\u03cd\u03bd\u03b5\u03b9 \u03bc\u03cc\u03bd\u03b7 \u03c4\u03b7\u03c2 \u03b1\u03c5\u03c4\u03ac \u03c4\u03b1 \u03c3\u03c5\u03c3\u03c4\u03b7\u03bc\u03b9\u03ba\u03ac \u03c0\u03c1\u03bf\u03b2\u03bb\u03ae\u03bc\u03b1\u03c4\u03b1.<\/p>\n<h2 id=\"tool-calling\">Native tool calling \u03bc\u03b5 \u03cc\u03c1\u03b9\u03b1<\/h2>\n<p>\u03a4\u03b1 Granite 4.2 \u03c5\u03c0\u03bf\u03c3\u03c4\u03b7\u03c1\u03af\u03b6\u03bf\u03c5\u03bd tool definitions \u03c3\u03c4\u03bf \u03b3\u03bd\u03c9\u03c3\u03c4\u03cc function-calling schema \u03ba\u03b1\u03b9 \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03c3\u03b5\u03c1\u03b2\u03b9\u03c1\u03b9\u03c3\u03c4\u03bf\u03cd\u03bd \u03b1\u03c0\u03cc vLLM \u03ae SGLang \u03bc\u03ad\u03c3\u03c9 OpenAI-compatible endpoint. \u03a4\u03bf model card \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03ac\u03c6\u03b5\u03b9 reasoning-augmented tool calling: \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b1\u03c0\u03bf\u03c6\u03b1\u03c3\u03af\u03b6\u03b5\u03b9 \u03c0\u03bf\u03b9\u03bf \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03b9 \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9 structured call \u03c0\u03bf\u03c5 \u03c4\u03bf harness \u03b5\u03ba\u03c4\u03b5\u03bb\u03b5\u03af. \u0397 \u03c3\u03c5\u03bc\u03b2\u03b1\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bc\u03b5\u03b9\u03ce\u03bd\u03b5\u03b9 \u03c4\u03bf integration work, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b5\u03bd \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03c4\u03bf output \u03c3\u03b5 \u03ad\u03bc\u03c0\u03b9\u03c3\u03c4\u03b7 \u03b5\u03bd\u03c4\u03bf\u03bb\u03ae.<\/p>\n<p>\u03a3\u03b5 \u03ad\u03bd\u03b1 e-commerce \u03ae customer-support workflow, \u03ba\u03ac\u03b8\u03b5 function \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c4\u03b5\u03bd\u03ae \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03cd\u03c3\u03b9\u03bc\u03b7. \u0397 \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7 stock, \u03b7 \u03b5\u03cd\u03c1\u03b5\u03c3\u03b7 \u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03af\u03c9\u03bd \u03c0\u03b1\u03c1\u03b1\u03b3\u03b3\u03b5\u03bb\u03af\u03b1\u03c2 \u03ba\u03b1\u03b9 \u03b7 \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03af\u03b1 \u03c0\u03c1\u03bf\u03c3\u03c7\u03b5\u03b4\u03af\u03bf\u03c5 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u03c2 \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03ad\u03c7\u03bf\u03c5\u03bd \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ac scopes. \u03a4\u03bf agent \u03b4\u03b5\u03bd \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b4\u03b9\u03ba\u03b1\u03af\u03c9\u03bc\u03b1 \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae\u03c2 \u03c4\u03b9\u03bc\u03ae\u03c2 \u03ae \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03bf\u03c6\u03ae\u03c2 \u03c7\u03c1\u03b7\u03bc\u03ac\u03c4\u03c9\u03bd \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b4\u03b9\u03b1\u03b2\u03ac\u03b6\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03c0\u03b1\u03c1\u03b1\u03b3\u03b3\u03b5\u03bb\u03af\u03b1.<\/p>\n<p>\u03a4\u03bf schema validation, \u03c4\u03b1 \u03b5\u03bb\u03ac\u03c7\u03b9\u03c3\u03c4\u03b1 permissions, \u03b7 \u03b5\u03c0\u03b9\u03b2\u03b5\u03b2\u03b1\u03af\u03c9\u03c3\u03b7 \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03bc\u03b7 \u03b1\u03bd\u03b1\u03c3\u03c4\u03c1\u03ad\u03c8\u03b9\u03bc\u03b7 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03bf audit trail \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03bf\u03c5\u03bd \u03b5\u03be\u03c9\u03c4\u03b5\u03c1\u03b9\u03ba\u03ac controls. \u03a4\u03bf \u03af\u03b4\u03b9\u03bf \u03b9\u03c3\u03c7\u03cd\u03b5\u03b9 \u03b3\u03b9\u03b1 prompt injection \u03bc\u03ad\u03c3\u03b1 \u03c3\u03b5 \u03b1\u03bd\u03b1\u03ba\u03c4\u03b7\u03bc\u03ad\u03bd\u03b1 \u03ad\u03b3\u03b3\u03c1\u03b1\u03c6\u03b1: \u03b7 \u03b1\u03bd\u03ac\u03bb\u03c5\u03c3\u03b7 \u03b3\u03b9\u03b1 \u03c4\u03b1 <a href=\"https:\/\/twodots.gr\/semantic-overlays-mi-ektelesimo-layer-prompt-injection\/\">Semantic Overlays \u03b1\u03c0\u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 \u03c3\u03c4\u03bf prompt injection<\/a> \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03b3\u03b9\u03b1\u03c4\u03af \u03c4\u03b1 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1 \u03c4\u03c1\u03af\u03c4\u03c9\u03bd \u03b4\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b1\u03c0\u03bf\u03ba\u03c4\u03bf\u03cd\u03bd \u03b5\u03be\u03bf\u03c5\u03c3\u03af\u03b1 \u03c0\u03ac\u03bd\u03c9 \u03c3\u03c4\u03b9\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03c4\u03bf\u03c5 agent.<\/p>\n<h2 id=\"benchmarks\">\u03a4\u03b9 \u03b4\u03b5\u03af\u03c7\u03bd\u03bf\u03c5\u03bd \u03c4\u03b1 benchmarks<\/h2>\n<p>\u03a3\u03c4\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03c0\u03bf\u03c5 \u03b4\u03b7\u03bc\u03bf\u03c3\u03af\u03b5\u03c5\u03c3\u03b5 \u03b7 IBM, \u03c4\u03bf 8B \u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 47,67 \u03c3\u03c4\u03bf SWE-Bench Verified \u03ba\u03b1\u03b9 20,56 \u03c3\u03c4\u03bf Terminal-Bench 2.1, \u03b5\u03bd\u03ce \u03c4\u03bf 30B \u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 57,00 \u03ba\u03b1\u03b9 29,24 \u03b1\u03bd\u03c4\u03af\u03c3\u03c4\u03bf\u03b9\u03c7\u03b1. \u03a3\u03c4\u03bf AIME25 \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03bf\u03bd\u03c4\u03b1\u03b9 78,33 \u03b3\u03b9\u03b1 \u03c4\u03bf 3B, 86,67 \u03b3\u03b9\u03b1 \u03c4\u03bf 8B \u03ba\u03b1\u03b9 89,17 \u03b3\u03b9\u03b1 \u03c4\u03bf 30B. \u039f\u03b9 \u03bc\u03b5\u03c4\u03c1\u03ae\u03c3\u03b5\u03b9\u03c2 \u03b4\u03b5\u03af\u03c7\u03bd\u03bf\u03c5\u03bd \u03c4\u03b7 \u03c3\u03c7\u03b5\u03c4\u03b9\u03ba\u03ae \u03ba\u03b1\u03c4\u03b5\u03cd\u03b8\u03c5\u03bd\u03c3\u03b7 \u03bc\u03ad\u03c3\u03b1 \u03c3\u03c4\u03b7\u03bd \u03bf\u03b9\u03ba\u03bf\u03b3\u03ad\u03bd\u03b5\u03b9\u03b1, \u03cc\u03c7\u03b9 \u03b5\u03b3\u03b3\u03c5\u03b7\u03bc\u03ad\u03bd\u03bf SLA \u03b3\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7.<\/p>\n<div class=\"td-chart td-chart--metrics\">\n<div class=\"td-chart-head\">\n<p class=\"td-chart-title\">\u03a4\u03ad\u03c3\u03c3\u03b5\u03c1\u03b1 \u03b4\u03b7\u03bc\u03bf\u03c3\u03b9\u03b5\u03c5\u03bc\u03ad\u03bd\u03b1 agentic \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1<\/p>\n<p class=\"td-chart-subtitle\">Vendor-reported pass@1 \u03c3\u03c4\u03b9\u03c2 \u03c1\u03c5\u03b8\u03bc\u03af\u03c3\u03b5\u03b9\u03c2 \u03c4\u03b7\u03c2 IBM\u00b7 \u03b1\u03c0\u03b1\u03b9\u03c4\u03b5\u03af\u03c4\u03b1\u03b9 \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03b7 \u03b4\u03bf\u03ba\u03b9\u03bc\u03ae \u03c3\u03c4\u03bf \u03b4\u03b9\u03ba\u03cc \u03c3\u03b1\u03c2 repository, terminal \u03ba\u03b1\u03b9 tool stack.<\/p>\n<\/div>\n<div class=\"td-chart-body\">\n<div class=\"td-metric-grid\">\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">47,67<\/span><span class=\"td-metric-label\">Granite 4.2 8B \u03c3\u03c4\u03bf SWE-Bench Verified<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">57,00<\/span><span class=\"td-metric-label\">Granite 4.2 30B \u03c3\u03c4\u03bf SWE-Bench Verified<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">20,56<\/span><span class=\"td-metric-label\">Granite 4.2 8B \u03c3\u03c4\u03bf Terminal-Bench 2.1<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">29,24<\/span><span class=\"td-metric-label\">Granite 4.2 30B \u03c3\u03c4\u03bf Terminal-Bench 2.1<\/span><\/div>\n<\/div>\n<\/div>\n<\/div>\n<p>\u03a4\u03b1 benchmarks \u03b4\u03b5\u03bd \u03b1\u03c0\u03b1\u03bd\u03c4\u03bf\u03cd\u03bd \u03b1\u03bd \u03c4\u03bf 30B \u03ad\u03c7\u03b5\u03b9 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03c3\u03c7\u03ad\u03c3\u03b7 \u03ba\u03cc\u03c3\u03c4\u03bf\u03c5\u03c2-\u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03bf\u03c2 \u03c3\u03b5 \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03cc support, CRM enrichment \u03ae marketing operations. \u0397 \u03b4\u03b7\u03bc\u03bf\u03c3\u03b9\u03b5\u03c5\u03bc\u03ad\u03bd\u03b7 \u03bb\u03af\u03c3\u03c4\u03b1 \u03b3\u03bb\u03c9\u03c3\u03c3\u03ce\u03bd \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 \u03b1\u03b3\u03b3\u03bb\u03b9\u03ba\u03ac, \u03b3\u03b5\u03c1\u03bc\u03b1\u03bd\u03b9\u03ba\u03ac, \u03b9\u03c3\u03c0\u03b1\u03bd\u03b9\u03ba\u03ac, \u03b3\u03b1\u03bb\u03bb\u03b9\u03ba\u03ac, \u03b9\u03b1\u03c0\u03c9\u03bd\u03b9\u03ba\u03ac, \u03c0\u03bf\u03c1\u03c4\u03bf\u03b3\u03b1\u03bb\u03b9\u03ba\u03ac, \u03b1\u03c1\u03b1\u03b2\u03b9\u03ba\u03ac, \u03c4\u03c3\u03b5\u03c7\u03b9\u03ba\u03ac, \u03b9\u03c4\u03b1\u03bb\u03b9\u03ba\u03ac, \u03ba\u03bf\u03c1\u03b5\u03b1\u03c4\u03b9\u03ba\u03ac, \u03bf\u03bb\u03bb\u03b1\u03bd\u03b4\u03b9\u03ba\u03ac \u03ba\u03b1\u03b9 \u03ba\u03b9\u03bd\u03b5\u03b6\u03b9\u03ba\u03ac, \u03cc\u03c7\u03b9 \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03ac. \u0386\u03c1\u03b1 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03be\u03b5\u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03cc \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03cc test set \u03b3\u03b9\u03b1 \u03ba\u03b1\u03c4\u03b1\u03bd\u03cc\u03b7\u03c3\u03b7, \u03c4\u03cc\u03bd\u03bf, \u03bf\u03c1\u03bf\u03bb\u03bf\u03b3\u03af\u03b1, tool arguments, \u03ac\u03c1\u03bd\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 recovery.<\/p>\n<aside class=\"td-article-note\"><strong>Benchmark \u2260 business outcome:<\/strong> \u03ad\u03bd\u03b1 \u03ba\u03b1\u03bb\u03cc coding score \u03b4\u03b5\u03bd \u03bc\u03b5\u03c4\u03c1\u03ac \u03c4\u03b7\u03bd \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c3\u03c4\u03b1 \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03ac, \u03c4\u03bf \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03b1\u03bd\u03ac \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03c9\u03bc\u03ad\u03bd\u03b7 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1, \u03c4\u03b1 \u03b4\u03b9\u03ba\u03b1\u03b9\u03ce\u03bc\u03b1\u03c4\u03b1 \u03b5\u03bd\u03cc\u03c2 tool \u03ae \u03c4\u03bf \u03c1\u03af\u03c3\u03ba\u03bf \u03bb\u03ac\u03b8\u03bf\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1\u03c2. \u03a3\u03c5\u03bd\u03b4\u03ad\u03c3\u03c4\u03b5 \u03ba\u03ac\u03b8\u03b5 model score \u03bc\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc acceptance test.<\/aside>\n<h2 id=\"deployment-quantization\">Deployment \u03ba\u03b1\u03b9 quantization<\/h2>\n<p>\u0397 \u03ac\u03b4\u03b5\u03b9\u03b1 Apache 2.0 \u03b5\u03c0\u03b9\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 download, fine-tuning \u03ba\u03b1\u03b9 \u03b5\u03bc\u03c0\u03bf\u03c1\u03b9\u03ba\u03ae \u03c7\u03c1\u03ae\u03c3\u03b7 \u03bc\u03b5 \u03c4\u03bf\u03c5\u03c2 \u03cc\u03c1\u03bf\u03c5\u03c2 \u03c4\u03b7\u03c2 \u03ac\u03b4\u03b5\u03b9\u03b1\u03c2. \u0397 IBM \u03c4\u03bf\u03c0\u03bf\u03b8\u03b5\u03c4\u03b5\u03af \u03c4\u03b7 \u03c3\u03b5\u03b9\u03c1\u03ac \u03c3\u03b5 cloud, on-premises \u03ba\u03b1\u03b9 edge \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd\u03c4\u03b1. \u0391\u03c5\u03c4\u03cc \u03b4\u03af\u03bd\u03b5\u03b9 \u03c3\u03c4\u03b9\u03c2 \u03bf\u03bc\u03ac\u03b4\u03b5\u03c2 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf \u03c0\u03ac\u03bd\u03c9 \u03c3\u03c4\u03b1 weights, \u03c3\u03c4\u03bf routing \u03c4\u03c9\u03bd \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03c9\u03bd \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b7\u03bd \u03c5\u03c0\u03bf\u03b4\u03bf\u03bc\u03ae, \u03b1\u03bb\u03bb\u03ac \u03bc\u03b5\u03c4\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03c3\u03b5 \u03b1\u03c5\u03c4\u03ad\u03c2 \u03c4\u03b7\u03bd \u03b5\u03c5\u03b8\u03cd\u03bd\u03b7 \u03b3\u03b9\u03b1 patching, monitoring, capacity planning \u03ba\u03b1\u03b9 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ae \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03af\u03b1.<\/p>\n<p>\u0393\u03b9\u03b1 inference \u03b4\u03b9\u03b1\u03c4\u03af\u03b8\u03b5\u03bd\u03c4\u03b1\u03b9 FP8, NVFP4, MXFP4 \u03ba\u03b1\u03b9 GGUF \u03c0\u03b1\u03c1\u03b1\u03bb\u03bb\u03b1\u03b3\u03ad\u03c2 \u03b1\u03c0\u03cc Q8_0 \u03ad\u03c9\u03c2 Q2_K. \u0397 FP8 \u03ad\u03ba\u03b4\u03bf\u03c3\u03b7 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af dynamic per-channel weights \u03ba\u03b1\u03b9 per-token activations \u03c7\u03c9\u03c1\u03af\u03c2 calibration. \u03a4\u03b1 NVFP4 \u03ba\u03b1\u03b9 MXFP4 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd GPTQ \u03bc\u03b5 2.000 SFT samples \u03ba\u03b1\u03b9 context 2K \u03ba\u03b1\u03c4\u03ac \u03c4\u03bf calibration. \u0391\u03c5\u03c4\u03ad\u03c2 \u03bf\u03b9 \u03bb\u03b5\u03c0\u03c4\u03bf\u03bc\u03ad\u03c1\u03b5\u03b9\u03b5\u03c2 \u03ad\u03c7\u03bf\u03c5\u03bd \u03c3\u03b7\u03bc\u03b1\u03c3\u03af\u03b1: \u03bc\u03b9\u03b1 \u03c3\u03c5\u03bc\u03c0\u03b9\u03b5\u03c3\u03bc\u03ad\u03bd\u03b7 \u03ad\u03ba\u03b4\u03bf\u03c3\u03b7 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b1\u03bb\u03bb\u03ac\u03be\u03b5\u03b9 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ac \u03c4\u03bf reasoning, \u03c4\u03b1 structured outputs \u03ba\u03b1\u03b9 \u03c4\u03b7\u03bd \u03b1\u03ba\u03c1\u03af\u03b2\u03b5\u03b9\u03b1 tool calls.<\/p>\n<p>\u03a4\u03bf self-hosting \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b4\u03c9\u03c1\u03b5\u03ac\u03bd \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03b4\u03b5\u03bd \u03c5\u03c0\u03ac\u03c1\u03c7\u03b5\u03b9 \u03ba\u03bb\u03b5\u03b9\u03c3\u03c4\u03cc API fee. \u039a\u03cc\u03c3\u03c4\u03bf\u03c2 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b7 GPU \u03ae CPU \u03c5\u03c0\u03bf\u03b4\u03bf\u03bc\u03ae, \u03b7 \u03bc\u03bd\u03ae\u03bc\u03b7, \u03b7 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1, \u03b7 \u03b4\u03b9\u03b1\u03b8\u03b5\u03c3\u03b9\u03bc\u03cc\u03c4\u03b7\u03c4\u03b1, \u03b7 observability, \u03bf\u03b9 \u03b1\u03bd\u03b1\u03b2\u03b1\u03b8\u03bc\u03af\u03c3\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 \u03bf \u03c7\u03c1\u03cc\u03bd\u03bf\u03c2 \u03c4\u03b7\u03c2 \u03bf\u03bc\u03ac\u03b4\u03b1\u03c2. \u0397 \u03c3\u03cd\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7 \u03b3\u03b9\u03b1 <a href=\"https:\/\/twodots.gr\/perissoteres-gpus-mikrotero-kv-cache-kostos-llm-serving\/\">\u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03b5\u03c2 GPUs \u03ae \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf KV cache<\/a> \u03b5\u03be\u03b7\u03b3\u03b5\u03af \u03b3\u03b9\u03b1\u03c4\u03af \u03c4\u03bf cost per completed task \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03b9\u03bf \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03b1\u03c0\u03bb\u03ae \u03c4\u03b9\u03bc\u03ae \u03b1\u03bd\u03ac token.<\/p>\n<h2 id=\"enterprise-governance\">\u03a4\u03b9 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ad\u03bd\u03b1 enterprise \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1<\/h2>\n<p>\u0388\u03bd\u03b1 \u03b1\u03bd\u03bf\u03b9\u03c7\u03c4\u03cc model \u03c0\u03c1\u03bf\u03c3\u03c6\u03ad\u03c1\u03b5\u03b9 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf, \u03cc\u03c7\u03b9 \u03ad\u03c4\u03bf\u03b9\u03bc\u03b7 \u03b4\u03b9\u03b1\u03ba\u03c5\u03b2\u03ad\u03c1\u03bd\u03b7\u03c3\u03b7. \u0397 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bf\u03c1\u03af\u03c3\u03b5\u03b9 \u03c0\u03bf\u03b9\u03b1 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1 \u03b5\u03c0\u03b9\u03c4\u03c1\u03ad\u03c0\u03b5\u03c4\u03b1\u03b9 \u03bd\u03b1 \u03b5\u03b9\u03c3\u03ad\u03c1\u03c7\u03bf\u03bd\u03c4\u03b1\u03b9, \u03c0\u03bf\u03b9\u03b1 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03b2\u03bb\u03ad\u03c0\u03b5\u03b9 \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf, \u03c0\u03bf\u03b9\u03b1 actions \u03b1\u03c0\u03b1\u03b9\u03c4\u03bf\u03cd\u03bd \u03ad\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7, \u03c0\u03cc\u03c3\u03bf \u03b4\u03b9\u03b1\u03c4\u03b7\u03c1\u03bf\u03cd\u03bd\u03c4\u03b1\u03b9 \u03c4\u03b1 logs \u03ba\u03b1\u03b9 \u03c0\u03bf\u03b9\u03bf\u03c2 \u03b1\u03bd\u03b1\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03b5\u03c5\u03b8\u03cd\u03bd\u03b7 \u03cc\u03c4\u03b1\u03bd \u03c4\u03bf \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bb\u03ac\u03b8\u03bf\u03c2. \u0391\u03c5\u03c4\u03ac \u03c4\u03b1 \u03cc\u03c1\u03b9\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bc\u03ad\u03c1\u03bf\u03c2 \u03c4\u03bf\u03c5 product design \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03c0\u03c1\u03cc\u03c3\u03b8\u03b5\u03c4\u03bf compliance layer \u03c3\u03c4\u03bf \u03c4\u03ad\u03bb\u03bf\u03c2.<\/p>\n<p>\u0397 \u03ba\u03bb\u03b9\u03bc\u03b1\u03ba\u03bf\u03cd\u03bc\u03b5\u03bd\u03b7 \u03b1\u03c5\u03c4\u03bf\u03bd\u03bf\u03bc\u03af\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c3\u03c4\u03b5\u03c1\u03b7 \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1 \u03b1\u03c0\u03cc\u03bb\u03c5\u03c4\u03bf \u00abagent \u03ae \u03cc\u03c7\u03b9\u00bb. \u039e\u03b5\u03ba\u03b9\u03bd\u03ae\u03c3\u03c4\u03b5 \u03bc\u03b5 read-only retrieval, \u03c0\u03b5\u03c1\u03ac\u03c3\u03c4\u03b5 \u03c3\u03b5 drafts, \u03bc\u03b5\u03c4\u03ac \u03c3\u03b5 \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03b5\u03c2 reversible actions \u03ba\u03b1\u03b9 \u03bc\u03cc\u03bd\u03bf \u03ad\u03c0\u03b5\u03b9\u03c4\u03b1 \u03c3\u03b5 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b5\u03be\u03bf\u03c5\u03c3\u03b9\u03bf\u03b4\u03cc\u03c4\u03b7\u03c3\u03b7, \u03b5\u03c6\u03cc\u03c3\u03bf\u03bd \u03c4\u03b1 evidence \u03b4\u03b5\u03af\u03c7\u03bd\u03bf\u03c5\u03bd \u03c3\u03c4\u03b1\u03b8\u03b5\u03c1\u03ae \u03b1\u03c0\u03cc\u03b4\u03bf\u03c3\u03b7. \u0397 \u03b1\u03bd\u03ac\u03bb\u03c5\u03c3\u03b7 \u03b3\u03b9\u03b1 <a href=\"https:\/\/twodots.gr\/ai-agents-me-oria-aftonomia-kai-krisi\/\">AI agents \u03bc\u03b5 \u03cc\u03c1\u03b9\u03b1 \u03ba\u03b1\u03b9 \u03ba\u03bb\u03b9\u03bc\u03b1\u03ba\u03bf\u03cd\u03bc\u03b5\u03bd\u03b7 \u03b1\u03c5\u03c4\u03bf\u03bd\u03bf\u03bc\u03af\u03b1<\/a> \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03b1\u03c5\u03c4\u03ae \u03c4\u03b7\u03bd \u03b1\u03c1\u03c7\u03ae \u03c3\u03b5 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b9\u03ba\u03cc \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf.<\/p>\n<p>\u03a4\u03bf NIST AI Risk Management Framework \u03bf\u03c1\u03b3\u03b1\u03bd\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7 \u03b4\u03bf\u03c5\u03bb\u03b5\u03b9\u03ac \u03b3\u03cd\u03c1\u03c9 \u03b1\u03c0\u03cc govern, map, measure \u03ba\u03b1\u03b9 manage. \u0393\u03b9\u03b1 Granite 4.2 \u03b1\u03c5\u03c4\u03cc \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03c3\u03b1\u03c6\u03ae use case, \u03b1\u03c0\u03bf\u03c4\u03cd\u03c0\u03c9\u03c3\u03b7 data flows \u03ba\u03b1\u03b9 permissions, \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7 \u03b1\u03bd\u03ac failure category, monitoring \u03bc\u03b5\u03c4\u03ac \u03c4\u03bf release \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03af\u03b1 \u03b1\u03bd\u03ac\u03ba\u03bb\u03b7\u03c3\u03b7\u03c2 \u03ae rollback. \u03a4\u03bf framework \u03b4\u03b5\u03bd \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03b9 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b3\u03b9\u03b1 \u03b5\u03c3\u03ac\u03c2\u00b7 \u03b2\u03bf\u03b7\u03b8\u03ac \u03bd\u03b1 \u03c4\u03b5\u03ba\u03bc\u03b7\u03c1\u03b9\u03ce\u03c3\u03b5\u03c4\u03b5 \u03b3\u03b9\u03b1\u03c4\u03af \u03ad\u03bd\u03b1 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf setup \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03bf\u03b4\u03b5\u03ba\u03c4\u03cc.<\/p>\n<div class=\"td-decision-band\">\n<p class=\"td-decision-kicker\">Go\/no-go \u03b3\u03b9\u03b1 Granite 4.2 agent<\/p>\n<p class=\"td-decision-title\">\u03a0\u03c1\u03bf\u03c7\u03c9\u03c1\u03ae\u03c3\u03c4\u03b5 \u03bc\u03cc\u03bd\u03bf \u03cc\u03c4\u03b1\u03bd model, runtime \u03ba\u03b1\u03b9 tools \u03c0\u03b5\u03c1\u03bd\u03bf\u03cd\u03bd \u03c4\u03bf \u03af\u03b4\u03b9\u03bf acceptance test<\/p>\n<p>Go \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03cc test set, \u03af\u03b4\u03b9\u03b1 prompts \u03ba\u03b1\u03b9 tools \u03b3\u03b9\u03b1 3B\/8B\/30B, \u03bc\u03b5\u03c4\u03c1\u03b7\u03bc\u03ad\u03bd\u03bf task success, \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2, latency, permissions, prompt-injection tests, approval gates, fallback \u03ba\u03b1\u03b9 rollback. No-go \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03b1\u03c0\u03cc benchmark table, \u03b1\u03bd\u03b5\u03be\u03ad\u03bb\u03b5\u03b3\u03ba\u03c4\u03b1 tools, \u03b1\u03c3\u03b1\u03c6\u03ad\u03c2 context limit \u03ae \u03b1\u03c5\u03c4\u03cc\u03bc\u03b1\u03c4\u03b7 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03ae \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1 \u03c7\u03c9\u03c1\u03af\u03c2 audit trail.<\/p>\n<\/div>\n<h2 id=\"pilot\">\u0395\u03c0\u03c4\u03ac \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03b3\u03b9\u03b1 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 pilot<\/h2>\n<p>\u0388\u03bd\u03b1 \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf pilot \u03b4\u03b5\u03bd \u03be\u03b5\u03ba\u03b9\u03bd\u03ac \u03bc\u03b5 \u00ab\u03c0\u03bf\u03b9\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b5\u03af\u03bd\u03b1\u03b9 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf;\u00bb, \u03b1\u03bb\u03bb\u03ac \u03bc\u03b5 \u03bc\u03b9\u03b1 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b7 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03ba\u03b1\u03b9 \u03bc\u03b5\u03c4\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf outcome. \u0393\u03b9\u03b1 coding agent \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 hidden tests \u03c0\u03bf\u03c5 \u03c0\u03b5\u03c1\u03bd\u03bf\u03cd\u03bd \u03c7\u03c9\u03c1\u03af\u03c2 \u03bc\u03b7 \u03b5\u03c0\u03b9\u03c4\u03c1\u03b5\u03c0\u03c4\u03ad\u03c2 \u03b1\u03bb\u03bb\u03b1\u03b3\u03ad\u03c2. \u0393\u03b9\u03b1 knowledge assistant \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 grounded \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03bc\u03b5 \u03c3\u03c9\u03c3\u03c4\u03ae \u03c0\u03b1\u03c1\u03b1\u03c0\u03bf\u03bc\u03c0\u03ae. \u0393\u03b9\u03b1 marketing operations \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 structured draft \u03c0\u03bf\u03c5 \u03b4\u03b5\u03bd \u03b4\u03b7\u03bc\u03bf\u03c3\u03b9\u03b5\u03cd\u03b5\u03c4\u03b1\u03b9 \u03c7\u03c9\u03c1\u03af\u03c2 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf.<\/p>\n<div class=\"td-step-list\">\n<p class=\"td-step-list-title\">\u0391\u03c0\u03cc \u03c4\u03bf checkpoint \u03c3\u03b5 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf enterprise workflow<\/p>\n<ol>\n<li><span class=\"td-step-kicker\">Schritt 1<\/span><strong>\u039f\u03c1\u03af\u03c3\u03c4\u03b5 \u03bc\u03af\u03b1 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03ba\u03b1\u03b9 \u03ad\u03bd\u03b1 acceptance test<\/strong>\n<p>\u03a3\u03c5\u03bb\u03bb\u03ad\u03be\u03c4\u03b5 30\u2013100 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac cases, \u03b1\u03c6\u03b1\u03b9\u03c1\u03ad\u03c3\u03c4\u03b5 \u03c0\u03c1\u03bf\u03c3\u03c9\u03c0\u03b9\u03ba\u03ac \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1 \u03cc\u03c0\u03bf\u03c5 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03b9 \u03b3\u03c1\u03ac\u03c8\u03c4\u03b5 \u03c4\u03b9 \u03b1\u03ba\u03c1\u03b9\u03b2\u03ce\u03c2 \u03b8\u03b5\u03c9\u03c1\u03b5\u03af\u03c4\u03b1\u03b9 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b1, \u03bc\u03b5\u03c1\u03b9\u03ba\u03ae \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b1 \u03ba\u03b1\u03b9 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Schritt 2<\/span><strong>\u039a\u03bb\u03b5\u03b9\u03b4\u03ce\u03c3\u03c4\u03b5 \u03ba\u03bf\u03b9\u03bd\u03ac prompts, tools \u03ba\u03b1\u03b9 permissions<\/strong>\n<p>\u0394\u03ce\u03c3\u03c4\u03b5 \u03c3\u03c4\u03b1 3B, 8B \u03ba\u03b1\u03b9 30B \u03c4\u03bf \u03af\u03b4\u03b9\u03bf tool schema, \u03c4\u03b1 \u03af\u03b4\u03b9\u03b1 \u03cc\u03c1\u03b9\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03bf \u03af\u03b4\u03b9\u03bf evaluation harness \u03ce\u03c3\u03c4\u03b5 \u03b7 \u03c3\u03cd\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7 \u03bd\u03b1 \u03b1\u03c6\u03bf\u03c1\u03ac \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ae \u03b5\u03bd\u03c3\u03c9\u03bc\u03ac\u03c4\u03c9\u03c3\u03b7.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Schritt 3<\/span><strong>\u039c\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 \u03c4\u03b1 \u03c4\u03c1\u03af\u03b1 reasoning modes \u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03ac<\/strong>\n<p>\u03a3\u03c5\u03b3\u03ba\u03c1\u03af\u03bd\u03b5\u03c4\u03b5 thinking, low-effort \u03ba\u03b1\u03b9 non-thinking \u03b3\u03b9\u03b1 task success, latency, tokens, invalid calls \u03ba\u03b1\u03b9 \u03b1\u03bd\u03ac\u03b3\u03ba\u03b7 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7\u03c2 \u03b4\u03b9\u03cc\u03c1\u03b8\u03c9\u03c3\u03b7\u03c2.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Schritt 4<\/span><strong>\u0394\u03bf\u03ba\u03b9\u03bc\u03ac\u03c3\u03c4\u03b5 \u03c4\u03bf \u03b1\u03ba\u03c1\u03b9\u03b2\u03ad\u03c2 deployment artifact<\/strong>\n<p>\u0395\u03bb\u03ad\u03b3\u03be\u03c4\u03b5 BF16 \u03ba\u03b1\u03b9 \u03c4\u03b7\u03bd quantized \u03ad\u03ba\u03b4\u03bf\u03c3\u03b7 \u03c0\u03bf\u03c5 \u03b8\u03b1 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b7\u03b8\u03b5\u03af, \u03c3\u03c4\u03bf \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc runtime, context length, concurrency \u03ba\u03b1\u03b9 hardware \u03c4\u03bf\u03c5 pilot.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0392\u03ae\u03bc\u03b1 5<\/span><strong>\u03a7\u03c4\u03c5\u03c0\u03ae\u03c3\u03c4\u03b5 \u03c4\u03b1 failure paths<\/strong>\n<p>\u0395\u03b9\u03c3\u03ac\u03b3\u03b5\u03c4\u03b5 timeout, \u03bb\u03ac\u03b8\u03bf\u03c2 tool output, \u03b5\u03bb\u03bb\u03b9\u03c0\u03ae \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1, prompt injection, \u03b1\u03bc\u03c6\u03af\u03c3\u03b7\u03bc\u03bf \u03b1\u03af\u03c4\u03b7\u03bc\u03b1 \u03ba\u03b1\u03b9 \u03bc\u03b7 \u03b5\u03c0\u03b9\u03c4\u03c1\u03b5\u03c0\u03c4\u03ae \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1 \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b9\u03b2\u03b5\u03b2\u03b1\u03b9\u03ce\u03c3\u03c4\u03b5 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 stop.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0392\u03ae\u03bc\u03b1 6<\/span><strong>\u03a5\u03c0\u03bf\u03bb\u03bf\u03b3\u03af\u03c3\u03c4\u03b5 \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03b1\u03bd\u03ac \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03c9\u03bc\u03ad\u03bd\u03b7 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1<\/strong>\n<p>\u03a3\u03c5\u03bd\u03c5\u03c0\u03bf\u03bb\u03bf\u03b3\u03af\u03c3\u03c4\u03b5 compute, \u03bc\u03bd\u03ae\u03bc\u03b7, retries, observability \u03ba\u03b1\u03b9 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03bf \u03c7\u03c1\u03cc\u03bd\u03bf \u03b1\u03bd\u03c4\u03af \u03bd\u03b1 \u03c3\u03c5\u03b3\u03ba\u03c1\u03af\u03bd\u03b5\u03c4\u03b5 \u03bc\u03cc\u03bd\u03bf tokens \u03ae \u03bc\u03ad\u03b3\u03b5\u03b8\u03bf\u03c2 checkpoint.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0392\u03ae\u03bc\u03b1 7<\/span><strong>\u039e\u03b5\u03ba\u03b9\u03bd\u03ae\u03c3\u03c4\u03b5 \u03c3\u03b5 shadow mode \u03bc\u03b5 rollback<\/strong>\n<p>\u039a\u03b1\u03c4\u03b1\u03b3\u03c1\u03ac\u03c8\u03c4\u03b5 \u03b1\u03c0\u03bf\u03c6\u03ac\u03c3\u03b5\u03b9\u03c2 \u03c7\u03c9\u03c1\u03af\u03c2 \u03b1\u03c5\u03c4\u03cc\u03bc\u03b1\u03c4\u03b7 \u03b5\u03c0\u03af\u03b4\u03c1\u03b1\u03c3\u03b7, \u03b5\u03b3\u03ba\u03c1\u03af\u03bd\u03b5\u03c4\u03b5 \u03c3\u03c4\u03b1\u03b4\u03b9\u03b1\u03ba\u03ac reversible actions \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03c4\u03b7\u03c1\u03ae\u03c3\u03c4\u03b5 deterministic fallback \u03ba\u03b1\u03b9 \u03c5\u03c0\u03b5\u03cd\u03b8\u03c5\u03bd\u03bf \u03b9\u03b4\u03b9\u03bf\u03ba\u03c4\u03ae\u03c4\u03b7.<\/p>\n<\/li>\n<\/ol>\n<\/div>\n<p>\u0397 \u03b1\u03be\u03af\u03b1 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03c1\u03b9\u03ad\u03c4\u03b1\u03b9 \u03c3\u03c4\u03b7\u03bd \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03c9\u03bc\u03ad\u03bd\u03b7 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03c3\u03c4\u03b7\u03bd \u03b5\u03bd\u03c4\u03cd\u03c0\u03c9\u03c3\u03b7 \u03b5\u03bd\u03cc\u03c2 demo. \u03a4\u03bf \u03c0\u03bb\u03b1\u03af\u03c3\u03b9\u03bf \u03b3\u03b9\u03b1 \u03c4\u03bf <a href=\"https:\/\/twodots.gr\/ai-roi-axia-ergasias\/\">AI ROI \u03ba\u03b1\u03b9 \u03c4\u03b7\u03bd \u03b1\u03be\u03af\u03b1 \u03c4\u03b7\u03c2 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1\u03c2<\/a> \u03b2\u03bf\u03b7\u03b8\u03ac \u03bd\u03b1 \u03c3\u03c5\u03bd\u03b4\u03b5\u03b8\u03bf\u03cd\u03bd task success, \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03bf\u03c2 \u03c7\u03c1\u03cc\u03bd\u03bf\u03c2, \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1.<\/p>\n<h2 id=\"symperasma\">\u0391\u03bd\u03bf\u03b9\u03c7\u03c4\u03cc \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf, \u03cc\u03c7\u03b9 \u03ad\u03c4\u03bf\u03b9\u03bc\u03b7 \u03b1\u03c5\u03c4\u03bf\u03bd\u03bf\u03bc\u03af\u03b1<\/h2>\n<p>\u03a4\u03bf IBM Granite 4.2 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b5\u03bd\u03b4\u03b9\u03b1\u03c6\u03ad\u03c1\u03bf\u03bd \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03c3\u03c5\u03bd\u03b4\u03c5\u03ac\u03b6\u03b5\u03b9 \u03c4\u03c1\u03af\u03b1 \u03bc\u03b5\u03b3\u03ad\u03b8\u03b7 dense reasoning models, \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03b1 thinking modes, native tool calling, Apache 2.0 \u03ba\u03b1\u03b9 agentic RL \u03c3\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac repositories, terminal sessions \u03ba\u03b1\u03b9 web search \u03b3\u03b9\u03b1 \u03c4\u03b1 8B \u03ba\u03b1\u03b9 30B. \u0397 \u03c4\u03b5\u03c7\u03bd\u03b9\u03ba\u03ae \u03b4\u03b9\u03b1\u03c6\u03ac\u03bd\u03b5\u03b9\u03b1 \u03b4\u03af\u03bd\u03b5\u03b9 \u03c3\u03c4\u03b9\u03c2 \u03bf\u03bc\u03ac\u03b4\u03b5\u03c2 \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03b1 \u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03af\u03b1 \u03b3\u03b9\u03b1 \u03bd\u03b1 \u03c3\u03c7\u03b5\u03b4\u03b9\u03ac\u03c3\u03bf\u03c5\u03bd \u03b4\u03bf\u03ba\u03b9\u03bc\u03ad\u03c2 \u03ba\u03b1\u03b9 deployment.<\/p>\n<p>\u0394\u03b5\u03bd \u03ba\u03b1\u03c4\u03b1\u03c1\u03b3\u03b5\u03af \u03c4\u03b7\u03bd \u03b1\u03bd\u03ac\u03b3\u03ba\u03b7 \u03b3\u03b9\u03b1 benchmarking, \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03ae \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7, governance \u03ae \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7 \u03b5\u03c0\u03af\u03b2\u03bb\u03b5\u03c8\u03b7. \u0397 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 3B \u03c7\u03c9\u03c1\u03af\u03c2 thinking \u03b3\u03b9\u03b1 \u03ad\u03bd\u03b1 \u03c3\u03c4\u03b5\u03bd\u03cc high-volume task, 8B \u03b3\u03b9\u03b1 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf agent \u03ae 30B \u03bc\u03cc\u03bd\u03bf \u03b3\u03b9\u03b1 \u03c4\u03b1 cases \u03cc\u03c0\u03bf\u03c5 \u03b1\u03c0\u03bf\u03b4\u03b5\u03b4\u03b5\u03b9\u03b3\u03bc\u03ad\u03bd\u03b1 \u03b1\u03c5\u03be\u03ac\u03bd\u03b5\u03b9 \u03c4\u03bf task success. \u03a4\u03bf \u03b6\u03b7\u03c4\u03bf\u03cd\u03bc\u03b5\u03bd\u03bf \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf, \u03b1\u03bb\u03bb\u03ac \u03c4\u03bf \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03c0\u03bf\u03c5 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03ce\u03bd\u03b5\u03b9 \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03b1 \u03c4\u03b7\u03bd \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03bc\u03ad\u03c3\u03b1 \u03c3\u03c4\u03b1 \u03b1\u03c0\u03bf\u03b4\u03b5\u03ba\u03c4\u03ac \u03cc\u03c1\u03b9\u03b1.<\/p>\n<section class=\"td-service-cta\">\n<div class=\"td-service-cta-content\">\n<p class=\"td-service-cta-kicker\">\u0391\u03c0\u03cc \u03c4\u03bf open model \u03c3\u03b5 \u03bc\u03b5\u03c4\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf workflow<\/p>\n<p class=\"td-service-cta-title\">\u03a3\u03c7\u03b5\u03b4\u03b9\u03ac\u03c3\u03c4\u03b5 AI agents \u03bc\u03b5 \u03c3\u03b1\u03c6\u03ae tools, approvals \u03ba\u03b1\u03b9 \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2<\/p>\n<p>\u0397 TWO DOTS \u03c7\u03b1\u03c1\u03c4\u03bf\u03b3\u03c1\u03b1\u03c6\u03b5\u03af use cases, model routing, permissions, test sets, observability \u03ba\u03b1\u03b9 fallback \u03ce\u03c3\u03c4\u03b5 \u03ad\u03bd\u03b1 Granite \u03ae \u03ac\u03bb\u03bb\u03bf self-hosted \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bd\u03b1 \u03b5\u03bd\u03c4\u03b1\u03c7\u03b8\u03b5\u03af \u03c3\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03af\u03b1 \u03c7\u03c9\u03c1\u03af\u03c2 \u03b1\u03bd\u03b5\u03be\u03ad\u03bb\u03b5\u03b3\u03ba\u03c4\u03b7 \u03b1\u03c5\u03c4\u03bf\u03bd\u03bf\u03bc\u03af\u03b1.<\/p>\n<div class=\"td-service-cta-actions\"><a class=\"td-service-cta-button\" href=\"https:\/\/twodots.gr\/aftomatismoi-epicheiriseon-ai\/\">\u0394\u03b5\u03af\u03c4\u03b5 \u0391\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2 \u0395\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03ae\u03c3\u03b5\u03c9\u03bd &amp; AI<\/a><\/div>\n<\/div>\n<\/section>\n<section id=\"sychnes-erotiseis\" class=\"td-faq-section\">\n<div class=\"td-faq\">\n<p class=\"td-faq-heading\">H\u00e4ufig gestellte Fragen<\/p>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a4\u03b9 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf IBM Granite 4.2;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0395\u03af\u03bd\u03b1\u03b9 \u03bf\u03b9\u03ba\u03bf\u03b3\u03ad\u03bd\u03b5\u03b9\u03b1 \u03b1\u03bd\u03bf\u03b9\u03c7\u03c4\u03ce\u03bd dense reasoning LLMs \u03c3\u03b5 3B, 8B \u03ba\u03b1\u03b9 30B \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03c4\u03c1\u03bf\u03c5\u03c2, \u03bc\u03b5 thinking modes, native tool calling \u03ba\u03b1\u03b9 \u03ac\u03b4\u03b5\u03b9\u03b1 Apache 2.0.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a5\u03c0\u03bf\u03c3\u03c4\u03b7\u03c1\u03af\u03b6\u03bf\u03c5\u03bd \u03cc\u03bb\u03b1 \u03c4\u03b1 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 thinking mode;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039d\u03b1\u03b9. \u039a\u03b1\u03b9 \u03c4\u03b1 \u03c4\u03c1\u03af\u03b1 \u03c5\u03c0\u03bf\u03c3\u03c4\u03b7\u03c1\u03af\u03b6\u03bf\u03c5\u03bd \u03c0\u03bb\u03ae\u03c1\u03b5\u03c2 thinking, non-thinking \u03ba\u03b1\u03b9 low-effort thinking \u03bc\u03ad\u03c3\u03c9 \u03c4\u03bf\u03c5 chat template, \u03ce\u03c3\u03c4\u03b5 \u03b7 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03bd\u03b1 \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03b9 reasoning budget \u03b1\u03bd\u03ac \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03bf\u03b9\u03b1 Granite 4.2 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 \u03b5\u03ba\u03c0\u03b1\u03b9\u03b4\u03b5\u03cd\u03c4\u03b7\u03ba\u03b1\u03bd \u03bc\u03b5 agentic RL;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a4\u03b1 8B \u03ba\u03b1\u03b9 30B \u03c0\u03ad\u03c1\u03b1\u03c3\u03b1\u03bd \u03c4\u03b1 \u03b5\u03be\u03b5\u03b9\u03b4\u03b9\u03ba\u03b5\u03c5\u03bc\u03ad\u03bd\u03b1 stages software engineering, terminal \u03ba\u03b1\u03b9 search. \u03a4\u03bf 3B \u03b1\u03ba\u03bf\u03bb\u03bf\u03cd\u03b8\u03b7\u03c3\u03b5 foundational RL \u03ba\u03b1\u03b9 alignment \u03c7\u03c9\u03c1\u03af\u03c2 \u03b1\u03c5\u03c4\u03cc \u03c4\u03bf agentic block.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03bf\u03b9\u03bf \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf context window \u03c4\u03bf\u03c5 Granite 4.2;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a4\u03b1 \u03b5\u03c0\u03af\u03c3\u03b7\u03bc\u03b1 \u03c5\u03bb\u03b9\u03ba\u03ac \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03bf\u03bb\u03cd\u03c4\u03c9\u03c2 \u03c3\u03c5\u03bd\u03b5\u03c0\u03ae: \u03c4\u03bf summary \u03ba\u03b1\u03b9 \u03c4\u03bf pre-training history \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03bf\u03c5\u03bd 512K, \u03b5\u03bd\u03ce \u03bf \u03c0\u03af\u03bd\u03b1\u03ba\u03b1\u03c2 \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae\u03c2 \u03ba\u03b1\u03b9 \u03c4\u03b1 serving examples \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd 131.072 tokens. \u0395\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03cd\u03c3\u03c4\u03b5 \u03c4\u03bf \u03cc\u03c1\u03b9\u03bf \u03c3\u03c4\u03bf \u03b1\u03ba\u03c1\u03b9\u03b2\u03ad\u03c2 checkpoint \u03ba\u03b1\u03b9 runtime.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a5\u03c0\u03bf\u03c3\u03c4\u03b7\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf Granite 4.2 \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03ac;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a4\u03b1 \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03ac \u03b4\u03b5\u03bd \u03b2\u03c1\u03af\u03c3\u03ba\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c3\u03c4\u03b7 \u03b4\u03b7\u03bc\u03bf\u03c3\u03b9\u03b5\u03c5\u03bc\u03ad\u03bd\u03b7 \u03bb\u03af\u03c3\u03c4\u03b1 \u03c5\u03c0\u03bf\u03c3\u03c4\u03b7\u03c1\u03b9\u03b6\u03cc\u03bc\u03b5\u03bd\u03c9\u03bd \u03b3\u03bb\u03c9\u03c3\u03c3\u03ce\u03bd. \u03a7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03bf \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03cc test set \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03ae \u03c7\u03c1\u03ae\u03c3\u03b7.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u039c\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b7\u03b8\u03b5\u03af \u03b5\u03bc\u03c0\u03bf\u03c1\u03b9\u03ba\u03ac;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a4\u03b1 Granite 4.2 LLMs \u03b4\u03b9\u03b1\u03c4\u03af\u03b8\u03b5\u03bd\u03c4\u03b1\u03b9 \u03bc\u03b5 Apache 2.0, \u03c0\u03bf\u03c5 \u03b5\u03c0\u03b9\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03b5\u03bc\u03c0\u03bf\u03c1\u03b9\u03ba\u03ae \u03c7\u03c1\u03ae\u03c3\u03b7, \u03c4\u03c1\u03bf\u03c0\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03bd\u03bf\u03bc\u03ae \u03bc\u03b5 \u03c4\u03bf\u03c5\u03c2 \u03cc\u03c1\u03bf\u03c5\u03c2 \u03c4\u03b7\u03c2 \u03ac\u03b4\u03b5\u03b9\u03b1\u03c2.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03bf\u03b9\u03b5\u03c2 quantized \u03b5\u03ba\u03b4\u03cc\u03c3\u03b5\u03b9\u03c2 \u03b4\u03b9\u03b1\u03c4\u03af\u03b8\u03b5\u03bd\u03c4\u03b1\u03b9;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0397 IBM \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 FP8, NVFP4, MXFP4 \u03ba\u03b1\u03b9 GGUF \u03c0\u03b1\u03c1\u03b1\u03bb\u03bb\u03b1\u03b3\u03ad\u03c2 \u03b1\u03c0\u03cc Q8_0 \u03ad\u03c9\u03c2 Q2_K. \u0397 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 reasoning \u03ba\u03b1\u03b9 tool calling \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03c0\u03b1\u03bd\u03b5\u03bb\u03b5\u03b3\u03c7\u03b8\u03b5\u03af \u03c3\u03c4\u03bf artifact \u03c0\u03bf\u03c5 \u03b8\u03b1 \u03b1\u03bd\u03b1\u03c0\u03c4\u03c5\u03c7\u03b8\u03b5\u03af.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03ce\u03c2 \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7 \u03b1\u03bd\u03ac\u03bc\u03b5\u03c3\u03b1 \u03c3\u03b5 3B, 8B \u03ba\u03b1\u03b9 30B;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039c\u03b5 \u03ba\u03bf\u03b9\u03bd\u03cc evaluation harness \u03ba\u03b1\u03b9 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac cases. \u03a3\u03c5\u03b3\u03ba\u03c1\u03af\u03bd\u03b5\u03b9 task success, latency, compute, tool-call accuracy, \u03b1\u03c0\u03cc\u03b4\u03bf\u03c3\u03b7 \u03c3\u03c4\u03b1 \u03b5\u03bb\u03bb\u03b7\u03bd\u03b9\u03ba\u03ac \u03ba\u03b1\u03b9 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7 \u03c0\u03b1\u03c1\u03ad\u03bc\u03b2\u03b1\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03b9 \u03c4\u03bf \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf setup \u03c0\u03bf\u03c5 \u03c0\u03b5\u03c1\u03bd\u03ac \u03c4\u03bf acceptance test.<\/p>\n<\/div>\n<\/details>\n<\/div>\n<\/section>\n<div class=\"td-source-list\">\n<p id=\"piges\" class=\"td-source-list-title\">Quellen<\/p>\n<ul>\n<li><a href=\"https:\/\/research.ibm.com\/blog\/introducing-granite-4-2\" target=\"_blank\" rel=\"noopener\">IBM Research \u2014 Granite 4.2 brings native reasoning to enterprise agents<\/a><\/li>\n<li><a href=\"https:\/\/huggingface.co\/blog\/ibm-granite\/granite-4-2\" target=\"_blank\" rel=\"noopener\">Granite Team, IBM \u2014 Granite 4.2 LLMs: How They\u2019re Built<\/a><\/li>\n<li><a href=\"https:\/\/huggingface.co\/ibm-granite\/granite-4.2-8b\" target=\"_blank\" rel=\"noopener\">IBM Granite \u2014 \u03b5\u03c0\u03af\u03c3\u03b7\u03bc\u03bf model card \u03c4\u03bf\u03c5 Granite 4.2 8B<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/ibm-granite\/granite-4.2-language-models\" target=\"_blank\" rel=\"noopener\">IBM Granite \u2014 \u03b5\u03c0\u03af\u03c3\u03b7\u03bc\u03bf repository \u03c4\u03c9\u03bd Granite 4.2 language models<\/a><\/li>\n<li><a href=\"https:\/\/www.nist.gov\/itl\/ai-risk-management-framework\" target=\"_blank\" rel=\"noopener\">NIST \u2014 Artificial Intelligence Risk Management Framework \u03ba\u03b1\u03b9 Generative AI Profile<\/a><\/li>\n<\/ul>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>\u03a4\u03bf IBM Granite 4.2 \u03c6\u03ad\u03c1\u03bd\u03b5\u03b9 reasoning, native tool calling \u03ba\u03b1\u03b9 \u03b1\u03bd\u03bf\u03b9\u03c7\u03c4\u03ac \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 3B, 8B \u03ba\u03b1\u03b9 30B \u03b3\u03b9\u03b1 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf\u03c5\u03c2 enterprise AI agents.<\/p>","protected":false},"author":1,"featured_media":106454,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_gspb_post_css":"","footnotes":""},"categories":[199],"tags":[6506,8768,21069,18293,21070],"class_list":["post-98388","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-techniti-noimosyni","tag-ai-agents","tag-enterprise-ai","tag-ibm-granite","tag-open-source-ai","tag-reasoning-models"],"blocksy_meta":{"styles_descriptor":{"styles":{"desktop":"","tablet":"","mobile":""},"google_fonts":[],"version":8}},"_links":{"self":[{"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/posts\/98388","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/comments?post=98388"}],"version-history":[{"count":2,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/posts\/98388\/revisions"}],"predecessor-version":[{"id":106457,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/posts\/98388\/revisions\/106457"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/media\/106454"}],"wp:attachment":[{"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/media?parent=98388"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/categories?post=98388"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/tags?post=98388"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}