MiniMax opens M3.1 Flash weights for coding and agents

A 428B MoE with 23B active parameters, 1M context and tunable thinking depth can now be self-hosted, with day-0 serving in two common inference stacks. Scores are MiniMax's own and unverified.

• context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
• attention — The mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.
• speculative decoding — A speed trick where a small model drafts several tokens ahead and the big model verifies them in one pass, often doubling generation speed.
• mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.

• postMiniMax M3.1 Flash: cheaper, faster agentic coding model, weights Oct 23
• postMiniMax Adds M3.1 Flash Preview to Existing Token Plan
• postMiniMax-H3 Runs 26x Faster on AMD GPUs via Partner Inference Stack
• postMiniMax Open-Sources Its Code CLI Under MIT License

• postMiniMax M3.1 Flash: cheaper, faster agentic coding model, weights Oct 23MiniMax (official)
• articleMiniMax M3.1 Flash releasedMiniMax model releases
• articleMiniMax M3.1 Flash lands on Together AI with 1M context and low pricingTogether AI

Checking sign-in…

Loading comments…