{"id":7858,"date":"2025-12-08T06:04:00","date_gmt":"2025-12-07T21:04:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=7858"},"modified":"2025-12-07T23:49:27","modified_gmt":"2025-12-07T14:49:27","slug":"mistral-3-deepseek-v3-2","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=7858","title":{"rendered":"Mistral 3, Deepseek V3.2, OpenRouter State of AI, Poetiq"},"content":{"rendered":"\n<p>\u5148\u9031\u306e\u5927\u304d\u306a\u8a71\u984c\u306fMistral3\u306e\u767a\u8868\uff08<a href=\"https:\/\/x.com\/MistralAI\/status\/1995872766177018340\">X\u30e6\u30fc\u30b6\u30fc\u306eMistral AI\u3055\u3093: \u300cIntroducing the Mistral 3 family of models: Frontier intelligence at all sizes. Apache 2.0. Details in \ud83e\uddf5 https:\/\/t.co\/lsrDmhW78u\u300d \/ X<\/a>\u3001<a href=\"https:\/\/mistral.ai\/news\/mistral-3\">Introducing Mistral 3 | Mistral AI<\/a>\uff09\u3068DeepSeek v3.2\uff08<a href=\"https:\/\/huggingface.co\/deepseek-ai\/DeepSeek-V3.2-Speciale\">deepseek-ai\/DeepSeek-V3.2-Speciale \u00b7 Hugging Face<\/a>\uff09\u3060\u3063\u305f\u3002\u3044\u305a\u308c\u3082\u5f37\u529b\u306a\u516c\u958b\u30e2\u30c7\u30eb\u3067\u3042\u308a\u3001\u30d5\u30ed\u30f3\u30c6\u30a3\u30a2\u30e2\u30c7\u30eb\u306b\u8fd1\u3044\u6027\u80fd\u3092\u4e3b\u5f35\u3057\u3066\u3044\u308b\u3002\u65b0\u305f\u306a\u30e2\u30c7\u30eb\u767a\u8868\u304c\u76f8\u6b21\u304e\u3001\u5b9f\u969b\u306e\u6027\u80fd\u691c\u8a3c\u306f\u3053\u308c\u304b\u3089\u3068\u3044\u3046\u611f\u3058\u3067\u306f\u3042\u308b\u304c\u3001\u7740\u5b9f\u306b\u7814\u7a76\u304c\u9032\u5c55\u3057\u3066\u3044\u308b\u611f\u304c\u3042\u308b\u3002<\/p>\n\n\n\n<p>OpenRouter\u304b\u3089\u767a\u8868\u3055\u308c\u305f<a href=\"https:\/\/openrouter.ai\/state-of-ai\">State of AI | OpenRouter<\/a>\u3082\u8208\u5473\u6df1\u3044\u30ec\u30dd\u30fc\u30c8\u3060\u3063\u305f\u3002\uff08\u30d0\u30a4\u30a2\u30b9\u306f\u3042\u308b\u306e\u3060\u308d\u3046\u304c\uff09\u30b3\u30fc\u30c9\u751f\u6210\u306b\u3088\u304f\u7528\u3044\u3089\u308c\u3066\u3044\u308b\u70b9\u3001\u591a\u69d8\u306a\u30e2\u30c7\u30eb\u304c\u5229\u7528\u3055\u308c\u3066\u3044\u308b\u70b9\u306a\u3069\u8208\u5473\u6df1\u3044\u3002<\/p>\n\n\n\n<p>Poetiq\u304b\u3089\u306eARC-AGI-2\u306eSoTA\uff08<a href=\"https:\/\/x.com\/poetiq_ai\/status\/1997027765393211881\">X\u30e6\u30fc\u30b6\u30fc\u306ePoetiq\u3055\u3093: \u300cPoetiq has officially shattered the ARC-AGI-2 SOTA \ud83d\ude80 @arcprize has officially verified our results: &#8211; 54% Accuracy \u2013 first to break the 50% barrier! &#8211; $30.57 \/ problem \u2013 less than half the cost of the previous best! We are now #1 on the leaderboard for ARC-AGI-2! https:\/\/t.co\/a8tPtCynVY\u300d \/ X<\/a>\uff09\u306b\u95a2\u3059\u308b\u767a\u8868\u3082\u8208\u5473\u6df1\u304b\u3063\u305f\u3002\u8a73\u7d30\u306a\u691c\u8a3c\uff08\u4ed6\u30c1\u30fc\u30e0\u306e\u3082\u306e\u3092\u542b\u3080\uff09\u5f85\u3061\u306e\u9762\u306f\u3042\u308b\u306e\u3060\u308d\u3046\u304c\u3001Agentic\u306a\u51e6\u7406\u3084\u8907\u6570\u306eLLM\u306e\u7d44\u307f\u5408\u308f\u305b\u306b\u306f\u73fe\u5728\u3067\u3082\u52b9\u679c\u304c\u3042\u308b\u3088\u3046\u306b\u601d\u3048\u308b\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Introducing Mistral 3<\/strong><br>Mistral 3\u304c\u767a\u8868\u3055\u308c\u300114B\u30018B\u30013B\u306e3\u7a2e\u985e\u306e\u5c0f\u578b\u30e2\u30c7\u30eb\u3068\u300141B\u306e\u30a2\u30af\u30c6\u30a3\u30d6\u30d1\u30e9\u30e1\u30fc\u30bf\u3092\u6301\u3064\u6700\u5f37\u306eMistral Large 3\u304c\u516c\u958b\u3055\u308c\u307e\u3057\u305f\u3002\u3053\u308c\u3089\u306f\u3059\u3079\u3066Apache 2.0\u30e9\u30a4\u30bb\u30f3\u30b9\u306e\u3082\u3068\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u5316\u3055\u308c\u3001\u958b\u767a\u8005\u30b3\u30df\u30e5\u30cb\u30c6\u30a3\u306b\u5f37\u3044\u57fa\u76e4\u3092\u63d0\u4f9b\u3057\u307e\u3059\u3002\u307e\u305f\u3001Mistral Large 3\u306f\u3001\u6700\u65b0\u306eNVIDIA GPU\u3092\u7528\u3044\u3066\u8a13\u7df4\u3055\u308c\u3001\u30de\u30eb\u30c1\u30e2\u30fc\u30c0\u30eb\u30fb\u30de\u30eb\u30c1\u30ea\u30f3\u30ac\u30eb\u51e6\u7406\u306b\u304a\u3044\u3066\u9ad8\u3044\u6027\u80fd\u3092\u767a\u63ee\u3057\u307e\u3059\u3002<\/li>\n\n\n\n<li><a href=\"https:\/\/mistral.ai\/news\/mistral-3\">Introducing Mistral 3 | Mistral AI<\/a><\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models&nbsp;<\/strong>[219.6]<br>\u672c\u7a3f\u3067\u306f,\u3088\u308a\u512a\u308c\u305f\u63a8\u8ad6\u3068\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u6027\u80fd\u3067\u9ad8\u3044\u8a08\u7b97\u52b9\u7387\u3092\u8abf\u548c\u3055\u305b\u308b\u30e2\u30c7\u30eb\u3067\u3042\u308bDeepSeek-V3.2\u3092\u7d39\u4ecb\u3059\u308b\u3002 \u8a08\u7b97\u8907\u96d1\u6027\u3092\u5927\u5e45\u306b\u4f4e\u6e1b\u3059\u308b\u52b9\u7387\u7684\u306a\u6ce8\u610f\u6a5f\u69cb\u3067\u3042\u308bDSA\u3092\u5c0e\u5165\u3059\u308b\u3002 DeepSeek-V3.2\u306f\u3001\u5805\u7262\u306a\u5f37\u5316\u5b66\u7fd2\u30d7\u30ed\u30c8\u30b3\u30eb\u3092\u5b9f\u88c5\u3057\u3001\u8a13\u7df4\u5f8c\u306e\u8a08\u7b97\u3092\u30b9\u30b1\u30fc\u30eb\u3059\u308b\u3053\u3068\u306b\u3088\u308a\u3001GPT-5\u3068\u540c\u7b49\u306b\u52d5\u4f5c\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2512.02556v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2512.02556v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Tue, 02 Dec 2025 09:25:14 GMT)<\/li>\n\n\n\n<li>DeepSeek Sparse Attention\u306a\u3069\u3001\u5185\u90e8\u69cb\u9020\u306b\u3082\u8e0f\u307f\u8fbc\u3093\u3060\u8ad6\u6587\u3002<\/li>\n\n\n\n<li>\u300c(3) Large-Scale Agentic Task Synthesis Pipeline: To integrate reasoning into tool-use scenarios, we developed a novel synthesis pipeline that systematically generates training data at scale. This methodology facilitates scalable agentic post-training, yielding substantial improvements in generalization and instruction-following robustness within complex, interactive environments.\u300d\u3068Agentic\u306a\u51e6\u7406\u5f37\u5316\u306b\u529b\u3092\u5165\u308c\u3066\u3044\u308b\u70b9\u306b\u3082\u6ce8\u76ee\u3002<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>State of AI | OpenRouter<\/strong><br>\u3053\u306e\u4e00\u5e74\u306f\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\uff08LLM\uff09\u306e\u9032\u5316\u3068\u5b9f\u4e16\u754c\u3067\u306e\u5229\u7528\u306b\u304a\u3044\u3066\u91cd\u8981\u306a\u8ee2\u6a5f\u3068\u306a\u3063\u305f\u30022024\u5e7412\u6708\u306b\u6700\u521d\u306e\u5e83\u304f\u63a1\u7528\u3055\u308c\u305f\u63a8\u8ad6\u30e2\u30c7\u30ebo1\u304c\u516c\u958b\u3055\u308c\u3001\u591a\u6bb5\u968e\u306e\u63a8\u8ad6\u304c\u53ef\u80fd\u306b\u306a\u308a\u3001\u958b\u767a\u3084\u5b9f\u9a13\u304c\u52a0\u901f\u3057\u305f\u3002\u5b9f\u969b\u306e\u4f7f\u7528\u306b\u95a2\u3059\u308b\u30c7\u30fc\u30bf\u5206\u6790\u306b\u3088\u308a\u3001\u30aa\u30fc\u30d7\u30f3\u30a6\u30a7\u30a4\u30c8\u30e2\u30c7\u30eb\u306e\u666e\u53ca\u3084\u5275\u9020\u7684\u306a\u30ed\u30fc\u30eb\u30d7\u30ec\u30a4\u306e\u4eba\u6c17\u306a\u3069\u3001\u591a\u69d8\u306a\u5229\u7528\u30d1\u30bf\u30fc\u30f3\u304c\u6d6e\u304b\u3073\u4e0a\u304c\u3063\u305f\u3002<\/li>\n\n\n\n<li><a href=\"https:\/\/openrouter.ai\/state-of-ai\">State of AI | OpenRouter<\/a><\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Poetiq Shatters ARC-AGI-2 State of the Art at Half the Cost<\/strong><br>Poetiq\u306f\u6700\u65b0\u306eGemini 3\u3068GPT-5.1\u30e2\u30c7\u30eb\u3092\u8fc5\u901f\u306b\u7d71\u5408\u3057\u3001\u4f4e\u30b3\u30b9\u30c8\u3067\u9ad8\u7cbe\u5ea6\u306a\u7d50\u679c\u3092\u5b9f\u73fe\u3057\u307e\u3057\u305f\u3002\u3053\u306e\u30b7\u30b9\u30c6\u30e0\u306f\u3001ARC-AGI-1\u304a\u3088\u3073ARC-AGI-2\u306b\u304a\u3044\u3066\u65b0\u3057\u3044\u30d1\u30ec\u30fc\u30c8\u30d5\u30ed\u30f3\u30c6\u30a3\u30a2\u3092\u7bc9\u304d\u3001\u5f93\u6765\u3092\u4e0a\u56de\u308b\u6027\u80fd\u3092\u767a\u63ee\u3057\u3066\u3044\u307e\u3059\u3002Poetiq\u306f\u307e\u305f\u3001\u904b\u7528\u306e\u67d4\u8edf\u6027\u3092\u6d3b\u304b\u3057\u3001\u6700\u9069\u306a\u30e2\u30c7\u30eb\u306e\u7d44\u307f\u5408\u308f\u305b\u3092\u81ea\u52d5\u7684\u306b\u9078\u629e\u3059\u308b\u80fd\u529b\u3092\u6301\u3064\u30e1\u30bf\u30b7\u30b9\u30c6\u30e0\u3092\u958b\u767a\u3057\u307e\u3057\u305f\u3002<\/li>\n\n\n\n<li><a href=\"https:\/\/mistral.ai\/news\/mistral-3\"><\/a><a href=\"https:\/\/poetiq.ai\/posts\/arcagi_verified\/\">Poetiq | ARC-AGI-2 SOTA at Half the Cost<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/poetiq.ai\/posts\/arcagi_announcement\/\">Poetiq | Traversing the Frontier of Superintelligence<\/a>\u306b\u3088\u308c\u3070\u300cThe prompt is an interface, not the intelligence: Our system engages in an iterative problem-solving loop. It doesn&#8217;t just ask a single question; it uses the LLM to generate a potential solution (sometimes code as in this example), receives feedback, analyzes the feedback, and then uses the LLM again to refine it. This multi-step, self-improving process allows us to incrementally build and perfect the answer.Self-Auditing: The system autonomously audits its own progress. It decides for itself when it has enough information and the solution is satisfactory, allowing it to terminate the process. This self-monitoring is critical for avoiding wasteful computation and minimizing costs.\u300d\u3068\u306e\u3053\u3068\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u304c\u516c\u958b\u3055\u308c\u3066\u3044\u308b\u3001<a href=\"https:\/\/github.com\/poetiq-ai\/poetiq-arc-agi-solver\">GitHub &#8211; poetiq-ai\/poetiq-arc-agi-solver: This repository allows reproduction of Poetiq&#8217;s record-breaking submission to the ARC-AGI-1 and ARC-AGI-2 benchmarks.<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u5148\u9031\u306e\u5927\u304d\u306a\u8a71\u984c\u306fMistral3\u306e\u767a\u8868\uff08X\u30e6\u30fc\u30b6\u30fc\u306eMistral AI\u3055\u3093: \u300cIntroducing the Mistral 3 family of models: Frontier intelligence at &hellip; <a href=\"https:\/\/devneko.jp\/wordpress\/?p=7858\" class=\"more-link\"><span class=\"screen-reader-text\">&#8220;Mistral 3, Deepseek V3.2, OpenRouter State of AI, Poetiq&#8221; \u306e<\/span>\u7d9a\u304d\u3092\u8aad\u3080<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[223],"class_list":["post-7858","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7858","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=7858"}],"version-history":[{"count":8,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7858\/revisions"}],"predecessor-version":[{"id":7903,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7858\/revisions\/7903"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7858"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7858"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7858"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}