# MiniMax ships M3, a Chinese open-weight model claiming frontier coding at one-twentieth the attention cost > A 1M-token sparse-attention model lands above GPT-5.5 on its own coding benchmark, below Claude Opus 4.8, with weights still withheld **Meta:** type: event · date: 2026-06-18 · heads: 長期戦, 語られていないこと · 7 takes · 5 lenses · 4 regions ## Summary [Chinese](/ja/entity/china) lab MiniMax released M3, an open-weight model pairing a 1M-token context window, native multimodal input and agentic computer use, and posted 59.0% on the SWE-Bench Pro coding benchmark, above OpenAI's GPT-5.5 (58.6%) and Google Gemini 3.1 Pro (54.2%) on the lab's own runs. It trails [Anthropic's](/ja/entity/anthropic) Claude Opus 4.8, shipped a week earlier, at a reported 69.2%. The headline engineering claim is MiniMax Sparse Attention (MSA), which selects only relevant key-value blocks and cuts per-token compute to one-twentieth at full context, with the architecture independently verified around June 18. The catch: the promised open weights had not been published at release, and training code and inference operators stayed closed. ## The split US ML press split between the capability story and the caveats. MarkTechPost foregrounded MSA's efficiency; Tech Times hammered that the benchmarks are vendor-run and that M3 sits below Opus 4.8. Outside the US, the framing shifted: Italy's developer coverage and India's Open Source For You centred two things US writeups soft-pedalled, that "open-weight" is not open-source with code withheld, and that China's 2017 National Intelligence Law obliges MiniMax to assist state intelligence on any prompt routed through its API. That governance angle, not the SWE-Bench number, is what the launch hype omits. ## By the numbers - 59.0%, M3's vendor-run SWE-Bench Pro score (GPT-5.5: 58.6%, Gemini 3.1 Pro: 54.2%). - 69.2%, Claude Opus 4.8's reported SWE-Bench Pro, ahead of M3. - 1M tokens, M3 context window. - 1/20, per-token compute at full context under MSA versus the prior generation. - 9x / 15x, faster prefill and decoding claimed under MSA. ## Why it matters Cheap long-context coding from an open-weight Chinese model pressures Western labs on price and pushes more inference toward Chinese infrastructure. But "open-weight" with withheld code, vendor benchmarks, and a legal duty to assist Beijing reframes the adoption question from capability to trust, especially for any team routing source code through the API. ## What to watch - Whether MiniMax actually publishes the M3 weights and a technical report. - Independent benchmark reruns versus the vendor numbers. - Enterprise and government bans or carve-outs over the API's data exposure. - Whether DeepSeek, [Qwen](/ja/entity/corporate/alibaba-qwen) or others match the sparse-attention efficiency. ## Regional takes (batched by bias / lens) ### ML engineering press - **MarkTechPost** (United States, en) — Technical writeup of M3's MiniMax Sparse Attention (MSA), which selects relevant key-value blocks to cut per-token compute to one-twentieth at 1M-token context, with native multimodal input and computer use for agentic coding. > "MSA cuts per-token compute to one-twentieth at 1M-token context, with over 9x faster prefill and 15x faster decoding than the prior generation." Source: https://www.marktechpost.com/2026/06/01/minimax-releases-minimax-m3-with-msa-architecture-supporting-1m-token-context-native-multimodality-and-agentic-coding/ ### skeptical benchmark scrutiny - **Tech Times** (United States, en) — Reports independent verification of the MSA architecture on June 18 while flagging that M3's 59.0% SWE-Bench Pro is vendor-run, that it trails Anthropic's Claude Opus 4.8 at 69.2%, and that promised open weights had not shipped. > "M3's 59.0% on SWE-Bench Pro beats GPT-5.5 but trails Claude Opus 4.8's 69.2%; the scores are company-run and the weights are still withheld." Source: https://www.techtimes.com/articles/318622/20260618/minimax-m3-takes-open-weight-ai-lead-sparse-attention-architecture-now-verified.htm ### European developer view - **Pasquale Pillitteri** (Italy, it) — European framing of M3 as a Chinese open-weights challenger to GPT-5.5, weighing the appeal of cheap long-context coding against China's 2017 National Intelligence Law obligations on any prompt sent to MiniMax's API. > "China's 2017 intelligence law obliges MiniMax to assist state intelligence work, an obligation that applies to every prompt sent to its API, wherever the user sits." Source: https://pasqualepillitteri.it/en/news/3934/minimax-m3-chinese-open-weights-coding-model ### Indian open-source community - **Open Source For You** (India, en) — Indian open-source view stressing that M3 is open-weight, not open-source: training code and inference operators stayed closed, so the 'open' label oversells what was actually released. > "M3 is open-weight, not open-source: MiniMax withheld training code and inference operators, stopping short of a full open commitment." Source: https://www.opensourceforu.com/2026/06/minimax-challenges-ai-rivals-with-m3-but-stops-short-of-full-open-source-commitment/ ### unlabelled - **TechCrunch** (United States, en) — Source: https://techcrunch.com/2026/06/01/minimax-m3/ - **DataNorth** (Netherlands, en) — Source: https://datanorth.ai/news/minimax-launches-m3 - **Kingy AI** (United States, en) — Source: https://kingy.ai/ai-launches/ai-launch-tracker-minimax-m3-specs-benchmarks-a-chinese-open-weight-model-takes-aim-at-the-frontier/ ## Across the graph - Related: [[anthropic-alibaba-claude-distillation]], [[china-dram-antitrust-probe-2026]] - Entities: China, Anthropic, Deepseek, Corporate:alibaba Qwen --- Canonical: https://rbtfl.xyz/ja/n/minimax-m3-open-weight-frontier-2026