MiniMax M3 Released with 1M Context and SOTA Coding Benchmarks
MiniMax_AI · x · 2026-08-27
MiniMax has released the M3 model, available on SambaCloud and designed for long-horizon agents. It features a 1M-token context window and introduces MiniMax Sparse Attention (MSA), delivering over 9x faster prefill and 15x faster decoding than its predecessor.
On benchmarks, M3 achieved 59.0% on SWE-Bench Pro, 66.0% on Terminal-Bench 2.1, and 74.2% on MCP Atlas. In internal testing, it autonomously optimized a CUDA kernel for 24 hours, boosting hardware peak utilization from 7.6% to 71.3%. The model is natively multimodal (text, image, video), capable of understanding charts and controlling desktops.
More from coding & agent
- Developer runs self-built Pokémon Emerald on original Game Boy Advance hardware — IanArawjo · 2026-08-27
- MARS: Multi-Specialist LLM Relay System Boosts Competitive Programming Pass Rates — Andrei Mikhailov · 2026-08-27
- Today's agents only think when pinged: a case for dissonance-driven cognitive initiative — GlenBradley · 2026-08-27
- Agent Architecture Recap: CoS Coordinating 20+ Agents and High-Volume PR Workflows — RachelVT42 · 2026-08-27
- Microsoft and Google's WebMCP standard adds agent buttons to websites — HankYeomans · 2026-08-27
- Offload Claude Code tasks to local Qwen model via MCP — CodeSlave9000 · 2026-08-27