MiniMax M3 M3.1 Flash
Released on October 19, 2026
Changes from previous version
Puts a public price and API id on the M3 Flash line after the September M3.1 Flash Preview. MiniMax M3 keeps the open weights and the $0.30/$1.20 list price.
Release Summary
MiniMax's public M3.1 Flash release, served as MiniMax-M3.1-Flash on the API, M Plan, and MiniMax Code. List price is $0.15 per million input tokens, $0.015 cached, and $0.60 output. A measured price per task is $0.22. On the announcement table, AutomationBench v1.0.6 is 62.8%, DeepSWE v1.1 is 73.7% with mini-SWE-agent 2.4.5, OSWorld 2.0 is 61.7%, Agents' Last Exam is 29.6% on the official Overall set, and CyberGym is 82.2% with Claude Code 2.1.263. The same chart labels DeepSWE at 72.1% for xhigh and 68.1% for high, and AutomationBench at 61.2% xhigh, 59.2% high, 57.7% medium, and 52.8% low under Claude Code. MiniMax says agent usage cost is 70% lower than MiniMax M3 and that everyday output speed can pass 200 tokens per second, with P50 above 170. The models.dev index lists a 1M-token context and a 512K output cap. Text and image in, text out. Weights come later, and MiniMax says a Pro model will follow. This release follows the Code-only M3.1 Flash Preview. Comparable rows on the same table: DeepSeek V4.1-Flash-1017, Kimi K3, GLM-5.3, GPT Sol 5.6, and Claude Opus 5.
Timeline
MiniMax M3.1 Flash released
MiniMax opens M3.1 Flash on the API, M Plan, and MiniMax Code, with weights planned for later.