MiniMax M3.1 Flash in practice: tests and comparisons with DeepSeek V4.1 Flash

OpenCode usage data for Sep 24–Oct 5 shows M3.1 Flash (formerly Space Bunny) first at 106T tokens, ahead of DeepSeek V4.1 Flash at 62T, Muse Spark 1.3 at 61T and MiMo V2.6 Flash at 16T, with a 3-day free period noted.

The post lists M3.1 Flash at $0.15 input and $0.60 output per million tokens, calling it one of MiniMax's cheapest models and slightly under DeepSeek V4.1 Flash.

A short note that MiniMax-M3.1-Flash open weights will be released on October 23.

Release breakdown: 428B MoE with ~23B active, 1M context, text/image/video input at $0.15/$0.60 per 1M tokens; reasoning can't be disabled and has five effort levels; 62.2 on Automation-Bench beats Opus 5 but Terminal Bench 4.0 trails by 13.7.

The author dropped DeepSeek from OpenCode after its latest mishap and now uses DSH, but plans to exploit OpenCode's free models to test whether M3.1 Flash lives up to its claims.

Pricing drops from M3's $0.30/$1.20 to a flat $0.15/$0.60, exactly matching DeepSeek 4.1 Flash 1017 off-peak rates, with a claimed 3x speedup and weights due October 23.

A side-by-side chart of the two same-week Flash releases covers price, coding, agent and vision: DeepSeek is stronger at code tasks while MiniMax stays competitive on agents.

Official specs roundup: 428B total / 23B active MoE, 1M context, thinking always on with Max default, and official scores trailing DeepSeek V4.1 Flash in 3 of 4 benchmarks.

The user says Space Bunny and M3.1-Preview performed badly on every task they tried and doubts the anonymous model was really M3.1 Flash.

The user says extensive testing of Space Bunny showed it underperforming GLM-5.3 Flash and DeepSeek V4.1 Flash in their workflow, and plans to retry M3.1 Flash on OpenCode without high expectations.

Input and output prices match DeepSeek, but cache reads at $0.015 are far pricier, widening the cost gap for cache-heavy workloads.

Stealth model Space Bunny is confirmed as an early M3.1 Flash, topping leaderboards with 200T+ tokens before anyone knew; $0.15/$0.60 per 1M tokens, DeepSWE 74%, open weights Oct 23.