MiniMax Releases M3 Open-Weights AI Model With 1 Million-Token Context, No. 6 Overall on Vals
MiniMax released M3, its first multimodal open-weights AI model aimed at coding and agentic tasks, with a 1 million-token context window. MiniMax said its Sparse Attention design makes that window practical by selecting blocks from an uncompressed KV cache, cutting the attention kernel's share of per-decode wall-clock time to about 5% from roughly 30% in the prior generation; the company has reported 59.0% on SWE-Bench Pro and 66.0% on Terminal Bench 2.1.
MiniMax said production serving with Together AI required work across paged decode, index scoring and multimodal preprocessing. The model is available through MiniMax's API and on services including SiliconFlow, where first-week rates were listed at $0.06, $0.30 and $1.20 per 1 million cached, input and output tokens. Vals ranked M3 as the top open-weights model on its Vals Index and Vals Multimodal Index and No. 6 overall, and said it led open-weights models on LegalBench, MedCode and Finance Agent V2, where it scored 48%, up 20 points from MiniMax-2.7. The weights have not yet been released, though MiniMax has said they will be published within about 10 days.
From the sources (25 posts)
@thdxrRT @opencode: MiniMax M3 will be launching soon You can try it right now in OpenCode For free
@scaling01MiniMax-M3 will come with a 1 million token context window
@minimax_aiRT @opencode: MiniMax M3 will be launching soon You can try it right now in OpenCode For free
@scaling01MiniMax-M3 is on OpenRouter - 1 million context - multi-modal - $0.6 / $2.4 per million tokens (input/output) currently 50% off
@scaling01MiniMax-M3 Benchmarks
@scaling01model weights and technical report in the next 10 days
@thezachmuellerRT @scaling01: MiniMax claims they scaled training data to the order of 100T tokens
@scaling01MiniMax claims they are now able to scale training data to the order of 100T tokens
@eliebakouchminimax M3 is the first open model trained on 100T+ tokens, natively multimodal, 1M context, built for long horizon tasks
@minimax_aiIntroducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities - Coding & Agentic Frontier: 59.0% SWE-Bench Pro, 66.0% Terminal Bench 2.1, 34.8% SWE-fficiency, 28.8% KernelBench Hard, 74.2% MCP Atlas - MiniMax
@stevibeRT @MiniMax_AI: Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities - Coding & Agentic Frontier: 5…
@minimax_ai🔥API Pricing & Promotion - 50% off standard usage (≤512K context) during the first 7 days - Priority access available through: api@minimax.io - Self-serve access for all users coming in the next few days
@omarsar0MiniMax M3 imminent. Will be doing deep testing with it on my own coding agent and harness. Review coming soon.
@minimax_aiRT @ollama: .@MiniMax_AI M3 model is available on Ollama's Cloud! In partnership with MiniMax, the M3 model on Ollama's Cloud is US-based…
@dair_aiRT @omarsar0: MiniMax M3 imminent. Will be doing deep testing with it on my own coding agent and harness. Review coming soon.
@arenaNew open model: MiniMax M3 by @MiniMax_AI is live in the Arena! Find it across Text, Vision, Document and Code Arena: Frontend. Bring your toughest prompts and vote. Scores incoming soon!
@togethercomputeMiniMax M3 is live and Together AI is powering its inference 🚀 Tomorrow at 6pm PT we're going live on X Spaces with the teams behind the model and the infrastructure to give you a deep dive.
@thezachmuellerRT @MiniMax_AI: Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities - Coding & Agentic Frontier: 5…
@minimax_aiRT @jiayuan_jy: 已经测试一个早上了，目前体感上接近 Opus 4.7（还需要进一步测试）。 用 M3 来写代码，Opus 4.8 + GPT5.5 来做对抗式的 code review，效果还不错。 已经完成了 1 个 PR
@calebfahlgrenRT @MiniMax_AI: Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities - Coding & Agentic Frontier: 5…
@eliebakouch> Weights & Tech Report in ~10 Days me:
@minimax_aiWe're going LIVE tomorrow with @togethercompute 🔥. @zpysky1125 is pulling back the curtain on M3: sparse attention, 1M context, all of it. You don't want to miss this.
@togethercomputeRT @MiniMax_AI: We're going LIVE tomorrow with @togethercompute 🔥. @zpysky1125 is pulling back the curtain on M3: sparse attention, 1M con…
@tekniumMinimax M3 is now live in Hermes Agent on Nous Portal, OpenRouter, and Minimax Direct Providers! No need to hermes update, it should appear in your model picker automatically!
@minimax_aiM3 on @OpenRouter same day we dropped it 🔥. 1M context, frontier coding + agentic, native multimodal. 50% off the first week.