Back to Model
M

MiniMax M3.1 Flash

Model Oct 19, 2026

MiniMax's October 19 hosted model for agent workflows, coding and computer use, with public token pricing and downloadable weights promised later.

Visit primary link 2 sources attached
MiniMax's Automation-Bench chart plots model scores against average cost per task on a logarithmic USD axis, with several reasoning settings per model.
MiniMax-published comparison from its October 19 M3.1 Flash announcement, not an AITrending benchmark. The source names Automation-Bench v1.0.6 and a Claude Code evaluation harness, with public model data as of October 18; its footer notes peak pricing for the DeepSeek comparison. Full frame preserved. Source: MiniMax
ai-agentscoding-agents

What happened lately

MiniMax M3.1 Flash timeline

MiniMax M3.1 Flash is a hosted language model announced on October 19, 2026. MiniMax identifies the previously anonymous Space Bunny as an early version. The company lists access through its API, M Plan and MiniMax Code, while saying model weights will arrive later.

The reviewed API documentation lists USD rates per million tokens of $0.15 input, $0.60 output and $0.015 cached input. Provider credentials and API billing apply. These token rates do not establish a fixed price for completing an agent task or the allowance included in a subscription.

MiniMax reports improvements on software engineering, computer-use and security benchmarks, plus median output speed above 170 tokens per second. These are provider-reported measurements; AITrending has not independently reproduced them.