MiniMax M3: Open-Weight Model with 1M Context Outperforms GPT-5.5 and Opus on Agentic Benchmarks
MiniMax has released M3, a strong open-weight model built for agentic tasks, featuring 1M-token context via MiniMax Sparse Attention and native multimodality. On benchmarks, M3 scores 59.0% on SWE-Bench Pro, 66.0% on Terminal Bench 2.1, 28.8% on KernelBench Hard, and 74.2% on MCP Atlas — outperforming both GPT-5.5 and Opus on BrowseComp, SVG Bench, and OSWorld Verified. Alongside the model, MiniMax also launched MiniMax Code, a dedicated coding environment available at code.minimax.io. API access is available at platform.minimax.io with a 50% discount on standard usage (up to 512K context) for the first 7 days.