MiniMax M3 Launches as Open-Weight AI Model Outperforming GPT-5.5 With 1M Token Context and Native Multimodal Support
Summary
MiniMax M3 launches as a groundbreaking open-weight AI model that outperforms GPT-5.5 and Gemini 2.5 Pro on key benchmarks, featuring a 1M token context window, native multimodal support, and autonomous capabilities powerful enough to independently reproduce a research paper over 12 hours, available now via desktop app and API starting at $20/month.
Key Points
- MiniMax M3 launches as the first open-weight model combining frontier-level coding performance, a 1M token context window powered by a new sparse attention architecture called MSA, and native multimodal support for image, video, and computer use.
- M3 surpasses GPT-5.5 and Gemini 3.1 Pro on key benchmarks including SWE-Bench Pro and Claw-Eval, and demonstrates exceptional long-horizon autonomous capabilities by independently reproducing an ICLR 2025 paper over 12 hours and optimizing a CUDA kernel by 9.4x over 24 hours.
- M3 is now accessible via the MiniMax Code desktop app, a tiered Token Plan starting at $20/month, and a developer API with flexible pricing based on input length and thinking mode, with model weights and a technical report set to be open-sourced within 10 days.