Chinese AI lab MiniMax releases M3 open-weight model with one million token context claiming SWE-Bench Pro coding scores close to Claude Opus
Chinese AI lab MiniMax released M3, an open-weight model with a one-million-token context window, claiming SWE-Bench Pro coding scores that beat GPT-5.5 and Gemini 3.1 Pro while approaching Claude Opus 4.7, according to The Decoder. Vendor benchmarks remain unverified.
Chinese AI lab MiniMax released M3 on June 1, positioning the system as the first open-weight model to combine frontier-class coding performance, a one-million-token context window and native multimodality, according to The Decoder. The company reports 59 percent on SWE-Bench Pro, which it claims beats GPT-5.5 and Gemini 3.1 Pro and approaches Claude Opus 4.7 from Anthropic. Pricing sits at roughly one tenth of Western closed-source competitors at launch. Several caveats follow. The benchmarks are vendor-reported and not yet independently verified. MiniMax compared M3 against Opus 4.7, but Anthropic shipped Opus 4.8 the week before M3's launch, which would tighten the gap on directly comparable agent evaluations. The promised open weights have not been released yet. Even with those qualifications, M3 marks another move by Chinese labs to compress the gap with US frontier model providers on both capability and price.