{"id":20930,"date":"2026-10-09T12:10:27","date_gmt":"2026-10-09T12:10:27","guid":{"rendered":"https:\/\/news.theck1.no\/?p=20930"},"modified":"2026-10-09T12:10:27","modified_gmt":"2026-10-09T12:10:27","slug":"mistrals-new-large-4-trails-some-chinese-open-models-in-independent-tests","status":"publish","type":"post","link":"https:\/\/news.theck1.no\/?p=20930","title":{"rendered":"Mistral\u2019s new Large 4 trails some Chinese open models in independent tests"},"content":{"rendered":"<p style=\"margin:0 0 1em; padding:0.6em 0.9em; border:1px solid #d0d7de; border-radius:6px; background:#f6f8fa; color:#444; font-size:0.9em;\"><strong>AI-rewritten:<\/strong> This is a summary of an article from Tom&#8217;s Hardware, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article &ndash; read it for the full story.<\/p>\n<div style=\"margin-bottom:1em; color:#666; font-size:0.9em;\"><strong>Tom&#8217;s Hardware &bull;  Shane Downing  &bull; October 9, 2026<\/strong><\/div>\n<hr\/>\n<p>Mistral recently launched its Large 4 (ML4) model, which uses a mixture-of-experts architecture trained on Nvidia&#8217;s Grace Blackwell GPUs in European datacenters. Artificial Analysis (AA), an independent benchmarking firm, ranked ML4 as the most intelligent open-weight model outside the U.S. and China with a score of 38 on its Intelligence Index v4.3.2. However, five Chinese models scored higher overall, including MiMo-V2.6-Pro at 46 and GLM-5.3 (max) at 45. Unlike these competitors, ML4 is not yet open; Mistral states its weights will be released by the end of October.<\/p>\n<p><!--more--><\/p>\n<p>In specific cybersecurity tests, ML4 achieved a personal best score of 81.7% on AA&#8217;s CyberGym-E2E-AA measure, surpassing GLM-5.3-Flash at 74%. CEO Arthur Mensch noted in Abu Dhabi that ML4 is superior to Chinese models in certain aspects like cyber security. Despite this strength, ML4 scored lower than competitors in other Cyber Index tests, such as CWE-Bench-AA at 51% and DeepsecBench-AA at 16%. Mistral claims the model will rank among the top three open weights on the Cyber Index once the weights ship.<\/p>\n<p>The cost per task for ML4 is $1.13 at list price or $0.57 at launch pricing, which is significantly higher than MiMo-V2.6-Pro at $0.13. Mistral trained ML4 from scratch using 3,800 Nvidia Grace Blackwell GPUs over two months, increasing its active parameters to 49B\u201352B compared to Large 3&#8217;s 41B. The model generates tokens at a rate of 116.1 per second and is reported to be 33% faster than its tier median. Mistral raised \u20ac3 billion in funding recently and plans to scale compute capacity further for future specialized models.<\/p>\n<div style=\"margin-top:2em; padding:1em; border-left:4px solid #0073aa; background:#f5f7fa;\">\n<p style=\"margin:0;\"><strong>Source:<\/strong> Tom&#8217;s Hardware &bull;  Shane Downing  &bull; October 9, 2026<\/p>\n<p style=\"margin:0.5em 0 0;\"><a href=\"https:\/\/www.tomshardware.com\/tech-industry\/artificial-intelligence\/independent-tests-rank-mistrals-new-trillion-parameter-large-4-the-best-ai-model-outside-the-u-s-and-china-but-chinese-open-weights-still-overcome-europes-best-efforts\" target=\"_blank\" rel=\"noopener\">Read the original article at Tom&#8217;s Hardware &rarr;<\/a><\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>AI-rewritten: This is a summary of an article from Tom&#8217;s Hardware, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article &ndash; read it for the full story. Tom&#8217;s Hardware &bull; Shane Downing &bull; October 9, 2026 Mistral recently launched its Large 4 (ML4) model, which<\/p>\n<p class=\"more-link\"><a href=\"https:\/\/news.theck1.no\/?p=20930\" class=\"themebutton2\">READ MORE<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7],"tags":[],"class_list":["post-20930","post","type-post","status-publish","format-standard","hentry","category-it-hardware"],"_links":{"self":[{"href":"https:\/\/news.theck1.no\/index.php?rest_route=\/wp\/v2\/posts\/20930","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/news.theck1.no\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/news.theck1.no\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/news.theck1.no\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/news.theck1.no\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=20930"}],"version-history":[{"count":0,"href":"https:\/\/news.theck1.no\/index.php?rest_route=\/wp\/v2\/posts\/20930\/revisions"}],"wp:attachment":[{"href":"https:\/\/news.theck1.no\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=20930"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/news.theck1.no\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=20930"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/news.theck1.no\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=20930"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}