Model Overview & Capabilities
MiniMax M3.1 Flash (officially launched October 19, 2026, following preview as ‘Space Bunny’) holds Rank #20 on our leaderboard with a composite score of 37.2. Built upon a 428-billion parameter Mixture-of-Experts architecture activating 23 billion parameters per token, it scores 37.5% on SWE-bench Verified, 37.0% on Terminal-Bench 2.0, 38.0% on Aider Polyglot, and 76.0% on GPQA Diamond with always-on internal reasoning.
MiniMax bills M3.1 Flash at $0.15 per million input tokens and $0.60 per million output tokens, giving an effective blended price of $0.26 per million tokens under 3:1 weighting and delivering a superior value rating of $0.0071 per index point. The model streams tokens at 181 tokens per second with an initial response latency of 0.70 seconds.
With a 1,000,000-token context window and multi-modal document understanding, MiniMax M3.1 Flash excels at computer use automation, complex spreadsheet and script execution, unit test generation, autonomous browser-based software testing tasks, and end-to-end continuous integration pipeline verification. Its active MoE routing ensures balanced computational density and minimal inference latency during heavy automated workflows.
