{"id":47470,"date":"2025-12-17T14:21:14","date_gmt":"2025-12-17T06:21:14","guid":{"rendered":"https:\/\/www.1ai.net\/?p=47470"},"modified":"2025-12-17T14:22:20","modified_gmt":"2025-12-17T06:22:20","slug":"%e5%b0%8f%e7%b1%b3%e7%aa%81%e7%84%b6%e5%8f%91%e5%b8%83%e6%96%b0%e6%a8%a1%e5%9e%8b%ef%bc%9a%e5%aa%b2%e7%be%8e-deepseek-v3-2%ef%bc%8c%e5%b0%8f%e7%b1%b3%e6%96%b0%e6%a8%a1%e5%9e%8b%e5%85%ac%e5%b8%83","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/47470.html","title":{"rendered":"Mi suddenly released a new model: DeepSeek-V3.2"},"content":{"rendered":"

December 17th message, just nowMillet<\/a>Officially released andOpen Source<\/a>newModel<\/a> MiMo-V2-Flash, first of all:<\/p>\n

\"e42ec5abj00t7ehke001xd000u000f0m\"<\/p>\n

MiMo-V2-Flash's total parameter 30.9 billion, active parameter 15 billion, using a mix of specialist structures (MoE) and performances that are also able to shake wrists with DeepSeek-V3.2 and Kimi-K2 head-open-source models\u3002<\/p>\n

Remove the \"open source\" label, MiMo-V2-Flash The real killer It's a radical innovation in architecture, pulling the pace of reasoning to 150 tokens\/seconds, at a cost of 0.1 dollars per million token, 0.3 dollars for output, and a super-extreme value\u3002<\/p>\n

The benchmark test results showed that MiMo-V2-Flash was in the top two of the open source models in both the AIME 2025 Math Competition and the GPQA-Diammond scientific knowledge test. The programming capability is even more bright, SWE-bench Verified score 73.4%, going beyond all open source models and pressing GPT-5-High\u3002<\/p>\n

Turning to intelligent missions, MiMo-V2-Flash scored 95.3, Retail 79.5, Aviation 66.0, Brownsecomp Search Agent 45.4, directly to 58.3 when context management was enabled\u3002<\/p>\n

These data show that MiMo-V2-Flash not only writes code, but also truly understands complex mission logic and implements multi-wheel intelligent interactions\u3002<\/p>\n

it is worth mentioning that mi used a hybrid slide window attention mechanism this time - using a 5-to-1 radical ratio, 5-sliding window attention paired with 1-slide global attention interchange, and the slide window looked at 128 tokens\u3002<\/p>\n

In response, Roosevelt pointed out on the social platform an intuitive discovery: window size 128 is the best dessert valueI don't know. It also stated that MiMo-V2-Flash was only the second step on the Mi-AGI road map\u3002<\/strong><\/p>\n

In addition, based on official experience page information, MiMo-V2-Flash supports in-depth thinking and networking search, both for chat and for use in scenes requiring real-time data, latest developments or data reconciliation\u3002<\/p>\n

In addition, MiMo-V2-Flash uses the MIT Open Source Protocol, and base copyright weights have also been published on Hugging Face\u3002<\/p>\n

Xiaomi Mimo AI Studio Experience Address: http:\/\/aistudio.xiaomimo.com<\/p>\n

HuggingFace Model Address: http:\/\/hf.co\/XiaomiMiMo\/MiMo-V2-Flash<\/p>\n

Technical coverage address: http:\/\/github.com\/XiaomiMiMo\/MiMo-V2-Flash\/blob\/main\/paper.pdf<\/p>","protected":false},"excerpt":{"rendered":"

On December 17th, a new MiMo-V2-Flash model was officially released and opened in Mimo-V2-Flash, first looking at performance: a total of 30.9 billion MiMo-V2-Flash, an active parameter of 15 billion, using an expert hybrid structure (MoE), and also able to bend wrists with these front-source models DeepSeek-V3.2 and Kimi-K2. Remove the \"open source\" label, MiMo-V2-Flash The real killer It's a radical innovation in architecture, pulling the pace of reasoning to 150 tokens\/seconds, at a cost of 0.1 dollars per million token, 0.3 dollars per output, and a super-extreme value. The benchmark test results were significant<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[1114,219,1489],"collection":[],"class_list":["post-47470","post","type-post","status-publish","format-standard","hentry","category-news","tag-1114","tag-219","tag-1489"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/47470","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=47470"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/47470\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=47470"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=47470"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=47470"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=47470"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}