Updated

MiniMax M3.1 Flash

MiniMax released M3.1 Flash for agentic tasks, coding, computer use, and cybersecurity. It is available through the MiniMax API, M Plan, and MiniMax Code; model weights are expected later, according to MiniMax.

By No Code MBA · Facts checked

Provider
MiniMax
Announcement date
Status at last check
Generally available
Model family
MiniMax M3
Preceding model
MiniMax-M3
Capabilities
Text, Coding

What changed

M3.1 Flash keeps MiniMax-M3's mixture-of-experts architecture (about 428B total, 23B active parameters) but adds sparse attention throughout, quantization-aware training and DSpark speculative decoding, with a 1M token context. MiniMax says online inference is nearly three times faster, at 200+ tokens per second in everyday use, and agentic task cost is 70% lower than the previous generation. MiniMax reports 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2 and 82% on CyberGym.

Access and availability

Available through the MiniMax API as MiniMax-M3.1-Flash, in the M Plan subscription, and in the MiniMax Code desktop app. A MiniMaxAI/MiniMax-M3.1-Flash repository is also listed on Hugging Face under the minimax-community license.

Official access documentation

Pricing

$0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens

Official pricing

What you can build

Test it on a bounded workflow involving several tool steps or a coding task, then compare task completion, latency, and human review needs with your current model.

Limitations

Benchmark and cybersecurity results are self-reported, and several comparisons lack named competitor scores. MiniMax says a more capable Pro model will arrive later, and its model card notes usable context and throughput depend on GPU memory and serving setup.

Sources and release history