MiniMax launches M3.1 Flash, a fast low-cost model for agent work

On October 19, 2026, MiniMax launched M3.1 Flash, which its release notes call an ultra-efficient open-weight model for everyday work, priced at $0.15 per million input tokens and $0.60 per million output tokens.

Archive entry: this covers an event from October 19, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On October 19, 2026, MiniMax launched MiniMax M3.1 Flash, which its release notes on the MiniMax platform call an ultra-efficient open-weight model for everyday work. The model was available via the API, M Plan and MiniMax Code.

What MiniMax reported

MiniMax reported nearly 3x faster inference and 70% lower agentic-task cost.
Output speed was 200+ tokens/s, with a P50 of 170+.
Price per million tokens was $0.15 for input, $0.60 for output and $0.015 for cached input.
Benchmarks: BankerToolBench 76.2, DeepSWE v1.1 73.7%, OSWorld V2 61.7, Agents' Last Exam 29.6 and CyberGym 82%.
An anonymous early version, Space Bunny, processed more than 200 trillion tokens.

What it means for people building with AI

Agent builders got a cheap, fast model for high-volume tool use, available in MiniMax Code and the API. The price was $0.15 per million input tokens and $0.60 per million output tokens, with cached input at $0.015. The speed, cost and benchmark figures were reported by MiniMax itself.

Weights and the Pro model

The MiniMax launch post said the model weights would be released later and that a Pro model was coming. When the MiniMaxAI/MiniMax-M3.1-Flash repository on Hugging Face was checked on November 7, 2026, it held weights of about 220B parameters under the minimax-community license.

Sources

Models (MiniMax, October 19, 2026, official)
MiniMax M3.1 Flash: Redefining Agent Performance and Cost - MiniMax Research | MiniMax (MiniMax, October 19, 2026, official)
MiniMaxAI/MiniMax-M3.1-Flash (MiniMax, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.