[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-minimax-m1-open-hybrid-attention-reasoning-model-en":3,"article-related-minimax-m1-open-hybrid-attention-reasoning-model-en":30,"series-model-release-6c57f6bf-1023-4a22-a6c0-013bd88ac3d1":83},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":22,"views":26,"created_at":27,"published_at":28,"topic_cluster_id":29},"6c57f6bf-1023-4a22-a6c0-013bd88ac3d1","minimax-m1-open-hybrid-attention-reasoning-model-en","MiniMax-M1 brings 1M-token open reasoning model","\u003Cp data-speakable=\"summary\">MiniMax-M1 is an open-source reasoning model with a 1 million-\u003Ca href=\"\u002Ftag\u002Ftoken\">token\u003C\u002Fa> context window.\u003C\u002Fp>\u003Cp>MiniMax unveiled \u003Ca href=\"https:\u002F\u002Fwww.minimax.io\u002Fnews\u002Fminimaxm1\" target=\"_blank\" rel=\"noopener\">MiniMax-M1\u003C\u002Fa> on June 16, 2025, and the headline numbers are hard to miss: a 1 million-token context window, 80,000-token reasoning output, and training that reportedly used 512 H800 GPUs for three weeks. The company says the full \u003Ca href=\"\u002Ftag\u002Freinforcement-learning\">reinforcement learning\u003C\u002Fa> phase cost $534,700, which is a very specific way to say this model was built to be efficient as well as large.\u003C\u002Fp>\u003Cp>For developers, the more interesting part is where M1 lands in practice. MiniMax says the model is open-source, tuned for productivity-heavy work, and already available through the \u003Ca href=\"https:\u002F\u002Fwww.minimax.io\u002F\" target=\"_blank\" rel=\"noopener\">MiniMax\u003C\u002Fa> app, web product, and API. It also ships with support from the open-source inference stack around \u003Ca href=\"https:\u002F\u002Fdocs.vllm.ai\u002F\" target=\"_blank\" rel=\"noopener\">vLLM\u003C\u002Fa> and \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fsgl-project\u002Fsglang\" target=\"_blank\" rel=\"noopener\">SGLang\u003C\u002Fa>, with weights and technical details on \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fminimax\" target=\"_blank\" rel=\"noopener\">Hugging Face\u003C\u002Fa> and \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FMiniMax-AI\" target=\"_blank\" rel=\"noopener\">GitHub\u003C\u002Fa>.\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Metric\u003C\u002Fth>\u003Cth>MiniMax-M1\u003C\u002Fth>\u003Cth>What it means\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>Context window\u003C\u002Ftd>\u003Ctd>1,000,000 tokens\u003C\u002Ftd>\u003Ctd>Matches Gemini 2.5 Pro and exceeds DeepSeek R1 by 8x\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Reasoning output\u003C\u002Ftd>\u003Ctd>80,000 tokens\u003C\u002Ftd>\u003Ctd>Long internal reasoning traces for complex tasks\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>RL training compute\u003C\u002Ftd>\u003Ctd>512 H800s for 3 weeks\u003C\u002Ftd>\u003Ctd>Reported reinforcement learning budget\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>RL rental cost\u003C\u002Ftd>\u003Ctd>$534,700\u003C\u002Ftd>\u003Ctd>MiniMax’s stated training cost\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>SWE-bench validation\u003C\u002Ftd>\u003Ctd>55.6% to 56.0%\u003C\u002Ftd>\u003Ctd>Strong software engineering benchmark result\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>API pricing\u003C\u002Ftd>\u003Ctd>$0.4 \u002F $2.2 per million tokens\u003C\u002Ftd>\u003Ctd>Input and output pricing for 0-200k tokens\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>Why MiniMax built M1 this way\u003C\u002Fh2>\u003Cp>MiniMax is betting that \u003Ca href=\"\u002Ftag\u002Flong-context\">long-context\u003C\u002Fa> reasoning is becoming a practical requirement, not a demo trick. If a model can keep a million tokens in working memory, it can hold large codebases, long documents, and extended tool traces without constant truncation. That matters for real workflows where the model has to read, think, revise, and keep track of prior steps.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778797872005-z8uk.png\" alt=\"MiniMax-M1 brings 1M-token open reasoning model\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>The company says M1 uses a hybrid-attention design with a \u003Ca href=\"https:\u002F\u002Fwww.minimax.io\u002Fnews\u002Fminimaxm1\" target=\"_blank\" rel=\"noopener\">Lightning Attention\u003C\u002Fa> mechanism. The pitch is simple: keep long-context computation efficient enough that the model can reason over huge inputs without burning absurd amounts of compute.\u003C\u002Fp>\u003Cp>That design choice also explains the training story. MiniMax says the reinforcement learning phase used a faster algorithm called CISPO, which clips importance sampling weights instead of relying on traditional token updates. The company claims this made convergence about twice as fast as other RL methods it compared against, including ByteDance’s DAPO.\u003C\u002Fp>\u003Cul>\u003Cli>1 million-token context window\u003C\u002Fli>\u003Cli>80,000-token reasoning output\u003C\u002Fli>\u003Cli>512 H800 GPUs used in RL training\u003C\u002Fli>\u003Cli>$534,700 reported RL rental cost\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>What the benchmarks actually say\u003C\u002Fh2>\u003Cp>MiniMax is careful to frame M1 as especially strong in software engineering, long-context understanding, and tool use. On \u003Ca href=\"https:\u002F\u002Fwww.swebench.com\u002F\" target=\"_blank\" rel=\"noopener\">SWE-bench\u003C\u002Fa> validation, the company reports 55.6% for MiniMax-M1-40k and 56.0% for MiniMax-M1-80k. That trails DeepSeek-R1-0528 at 57.6%, but it still puts M1 ahead of other open-weight models in the company’s comparisons.\u003C\u002Fp>\u003Cp>The more eye-catching claim is long-context performance. MiniMax says the M1 series beats all open-weight models on long-context understanding and even ranks above \u003Ca href=\"https:\u002F\u002Fopenai.com\u002Findex\u002Fo3\u002F\" target=\"_blank\" rel=\"noopener\">OpenAI o3\u003C\u002Fa> and \u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fnews\u002Fclaude-4\" target=\"_blank\" rel=\"noopener\">Claude 4 Opus\u003C\u002Fa>, landing second overall behind \u003Ca href=\"https:\u002F\u002Fdeepmind.google\u002Ftechnologies\u002Fgemini\u002Fpro\u002F\" target=\"_blank\" rel=\"noopener\">Gemini 2.5 Pro\u003C\u002Fa>. In \u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa> tool-use tests on TAU-bench, MiniMax says M1-40k beats every open-weight model and also tops Gemini 2.5 Pro.\u003C\u002Fp>\u003Cblockquote>\"This feature gives us a substantial computational efficiency advantage in both training and inference.\" — MiniMax\u003C\u002Fblockquote>\u003Cp>That quote matters because it gets to the real business of the release. A giant context window is impressive, but if it costs too much to train or run, it stays a lab curiosity. MiniMax is trying to argue the opposite: that M1 is large enough for serious work and cheap enough to ship widely.\u003C\u002Fp>\u003Ch2>Price is part of the product here\u003C\u002Fh2>\u003Cp>MiniMax is not treating pricing as an afterthought. The company says M1 is free to use in the MiniMax app and on the web, and its \u003Ca href=\"\u002Fnews\u002Fwhy-openai-api-pricing-is-product-strategy-en\">API pricing\u003C\u002Fa> is aimed at undercutting higher-end rivals. For inputs between 0 and 200k tokens, the price is $0.4 per million input tokens and $2.2 per million output tokens. For 200k to 1M token inputs, the price rises to $1.3 per million input tokens, while output stays at $2.2 per million tokens.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778797859045-3rol.png\" alt=\"MiniMax-M1 brings 1M-token open reasoning model\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>That pricing structure matters because long-context models usually get expensive fast. MiniMax is signaling that it wants developers to test huge prompts, long codebases, and extended agent traces without treating every run like a budget decision.\u003C\u002Fp>\u003Cul>\u003Cli>0-200k input: $0.4 per million tokens\u003C\u002Fli>\u003Cli>0-200k output: $2.2 per million tokens\u003C\u002Fli>\u003Cli>200k-1M input: $1.3 per million tokens\u003C\u002Fli>\u003Cli>200k-1M output: $2.2 per million tokens\u003C\u002Fli>\u003C\u002Ful>\u003Cp>There is also a practical ecosystem angle. MiniMax says the model is already supported by \u003Ca href=\"https:\u002F\u002Fdocs.vllm.ai\u002F\" target=\"_blank\" rel=\"noopener\">vLLM\u003C\u002Fa> and \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fsgl-project\u002Fsglang\" target=\"_blank\" rel=\"noopener\">SGLang\u003C\u002Fa>, which matters because teams do not want to wait months for tooling to catch up. If a model is open but hard to deploy, adoption slows down fast.\u003C\u002Fp>\u003Ch2>What developers should watch next\u003C\u002Fh2>\u003Cp>M1 is a strong signal that the open-weight race is moving from raw parameter bragging to practical constraints like context length, inference cost, and agent performance. MiniMax is trying to win on all three at once: open weights, million-token memory, and API pricing that invites experimentation.\u003C\u002Fp>\u003Cp>The real test is whether teams use M1 for tasks that punish weaker models: large codebase refactors, long document analysis, multi-step tool workflows, and agent loops that need to remember what happened 50,000 tokens ago. If MiniMax’s claims hold up outside its own report, M1 could become a serious option for developers who care more about throughput and context than brand names.\u003C\u002Fp>\u003Cp>MiniMax also said more updates are coming over the next four workdays, so this release may be the first move in a larger product push. The question now is simple: can M1 keep its \u003Ca href=\"\u002Ftag\u002Fbenchmark\">benchmark\u003C\u002Fa> edge once more teams run it on their own workloads, with their own prompts, costs, and failure cases?\u003C\u002Fp>\u003Cp>If you want the practical takeaway, it is this: watch the open-source models that can handle very long contexts without punishing your GPU bill. That is where the next wave of useful reasoning systems is likely to get judged.\u003C\u002Fp>","MiniMax released M1, an open-source reasoning model with 1M-token context, 80k output, and low-cost API pricing.","www.minimax.io","https:\u002F\u002Fwww.minimax.io\u002Fnews\u002Fminimaxm1",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778797872005-z8uk.png","model-release","en","5b5fa24f-5259-4e9e-8270-b08b6805f281",[17,18,19,20,21],"MiniMax-M1","open-source AI","hybrid attention","long context","reasoning model",[23,24,25],"MiniMax-M1 is an open-source reasoning model with a 1 million-token context window and 80,000-token output.","MiniMax says the RL phase used 512 H800s for three weeks at a cost of $534,700.","The model is priced aggressively and is already available through MiniMax’s app, web, and API.",11,"2026-05-14T22:30:39.599473+00:00","2026-05-14T22:30:39.585+00:00","8a720a1b-e905-4cc6-8607-4887b319116e",{"tags":31,"relatedLang":42,"relatedPosts":46},[32,34,36,38,40],{"name":19,"slug":33},"hybrid-attention",{"name":17,"slug":35},"minimax-m1",{"name":18,"slug":37},"open-source-ai",{"name":20,"slug":39},"long-context",{"name":21,"slug":41},"reasoning-model",{"id":15,"slug":43,"title":44,"language":45},"minimax-m1-open-hybrid-attention-reasoning-model-zh","MiniMax-M1：開源 1M Token 推理模型","zh",[47,53,59,65,71,77],{"id":48,"slug":49,"title":50,"cover_image":51,"image_url":51,"created_at":52,"category":13},"58aa41ca-2c5f-44c6-ab07-2002473e95b1","gemini-1-5-pro-002-flash-002-2-0-flash-update-en","Gemini 1.5 Pro-002, Flash-002 and 2.0 Flash update Google AI","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780999383257-jccn.png","2026-06-09T10:02:28.362637+00:00",{"id":54,"slug":55,"title":56,"cover_image":57,"image_url":57,"created_at":58,"category":13},"435fc551-a461-444a-bf95-dbf5685cfac0","minimax-m3-open-weight-coding-win-en","MiniMax M3 Proves Open-Weight Can Still Win on Coding","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780968781159-odhi.png","2026-06-09T01:32:31.256895+00:00",{"id":60,"slug":61,"title":62,"cover_image":63,"image_url":63,"created_at":64,"category":13},"12af5a0d-1bbf-4a50-a391-b53f8003f234","gemini-35-flash-pricing-benchmarks-en","Gemini 3.5 Flash Pricing, Context, Benchmarks","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780840981235-e7hm.png","2026-06-07T14:02:30.280485+00:00",{"id":66,"slug":67,"title":68,"cover_image":69,"image_url":69,"created_at":70,"category":13},"0e767e9d-5d17-4cd0-b6ee-0328f89eb49b","gemma-4-12b-specs-benchmarks-run-locally-en","Gemma 4 12B: Specs, Benchmarks & How to Run It Locally","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780777984661-5ymr.png","2026-06-06T20:32:25.294996+00:00",{"id":72,"slug":73,"title":74,"cover_image":75,"image_url":75,"created_at":76,"category":13},"9d15f962-739d-44f8-a7f9-11bca64d38e0","best-kimi-models-2026-k2-5-vs-k2-thinking-en","Best Kimi Models in 2026: K2.5 vs K2 Thinking","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780770786284-shy0.png","2026-06-06T18:32:39.779504+00:00",{"id":78,"slug":79,"title":80,"cover_image":81,"image_url":81,"created_at":82,"category":13},"34547376-5d6b-4453-8d80-8072d8ac36ed","kimi-k2-6-open-source-coding-agent-swarm-en","Kimi K2.6 adds open-source coding and agent swarm","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780761781526-wop4.png","2026-06-06T16:02:22.26883+00:00",[84,89,94,99,104,109,114,119,124,129],{"id":85,"slug":86,"title":87,"created_at":88},"d4cffde7-9b50-4cc7-bb68-8bc9e3b15477","nvidia-rubin-ai-supercomputer-en","NVIDIA Unveils Rubin: A Leap in AI Supercomputing","2026-03-25T16:24:35.155565+00:00",{"id":90,"slug":91,"title":92,"created_at":93},"eab919b9-fbac-4048-89fc-afad6749ccef","google-gemini-ai-innovations-2026-en","Google's AI Leap with Gemini Innovations in 2026","2026-03-25T16:27:18.841838+00:00",{"id":95,"slug":96,"title":97,"created_at":98},"5f5cfc67-3384-4816-a8f6-19e44d90113d","gap-google-gemini-ai-checkout-en","Gap Teams Up with Google Gemini for AI-Driven Checkout","2026-03-25T16:27:46.483272+00:00",{"id":100,"slug":101,"title":102,"created_at":103},"f6d04567-47f6-49ec-804c-52e61ab91225","ai-model-release-wave-march-2026-en","Navigating the AI Model Release Wave of March 2026","2026-03-25T16:28:45.409716+00:00",{"id":105,"slug":106,"title":107,"created_at":108},"895c150c-569e-4fdf-939d-dade785c990e","small-language-models-transform-ai-en","Small Language Models: Llama 3.2 and Phi-3 Transform AI","2026-03-25T16:30:26.688313+00:00",{"id":110,"slug":111,"title":112,"created_at":113},"38eb1d26-d961-4fd3-ae12-9c4089680f5f","midjourney-v8-alpha-features-pricing-en","Midjourney V8 Alpha: A Deep Dive into Its Features and Pricing","2026-03-26T01:25:36.387587+00:00",{"id":115,"slug":116,"title":117,"created_at":118},"bf36bb9e-3444-4fb8-ab19-0df6bc9d8271","rag-2026-indispensable-ai-bridge-en","RAG in 2026: The Indispensable AI Bridge","2026-03-26T01:28:34.472046+00:00",{"id":120,"slug":121,"title":122,"created_at":123},"60881d6d-2310-44ef-b1fb-7f98e9dd2f0e","xiaomi-mimo-trio-agents-robots-voice-en","Xiaomi’s MiMo trio targets agents, robots, and voice","2026-03-28T03:05:08.899895+00:00",{"id":125,"slug":126,"title":127,"created_at":128},"f063d8d1-41d1-4de4-8ebc-6c40511b9369","xiaomi-mimo-v2-pro-1t-moe-agents-en","Xiaomi MiMo-V2-Pro: 1T MoE Model for Agents","2026-03-28T03:06:19.238032+00:00",{"id":130,"slug":131,"title":132,"created_at":133},"a1379e9a-6785-4ff5-9b0a-8cff55f8264f","cursor-composer-2-started-from-kimi-en","Cursor’s Composer 2 started from Kimi","2026-03-28T03:11:59.132398+00:00"]