M3.1 Flash
A lightweight, high-speed large language model released by MiniMax, designed for fast, low-latency inference in high-concurrency and latency-sensitive business scenarios.
Core Facts
M3.1 Flash can be called via the AI SDK, Chat Completions, Responses, and the Anthropic Messages API, with the model name minimax/minimax-m3.1-flash.
The official name of MiniMax M3.1 Flash is `MiniMax-M3.1-Flash`, positioned as a frontier multimodal coding model with a 1 million context window and adjustable thinking depth.
MiniMax M3.1 Flash is available on AIHubMix for token-billed calls under the model ID `minimax-m3.1-flash`, with input priced at $0.165 per million tokens and output at $0.66 per million tokens.
Timeline (last 90 days)
vLLM provided Day-0 support for MiniMax M3.1 Flash on the day it was released.
vLLM integrated three serving-stack configurations for M3.1 Flash: native NVFP4 KV cache, NVFP4 MoE weights, and DSpark speculative decoding.
Whether a model can be delivered by a host under an SLA is an important signal of its engineering maturity.
Engineering improvements related to MiniMax M3.1 Flash are being upstreamed to vLLM.
@inferact provides SLA-guaranteed inference hosting services for MiniMax M3.1 Flash.
M3.1 Flash's output speed reaches 200+ tokens per second.
M3.1 Flash reduces costs on agent tasks by 70%.
M3.1 Flash's online speed is about 3 times that of the previous generation.
The officially stated 3x speed and 70% cost reduction are vendor-supplied figures, and actual performance depends on task type, context length and concurrency conditions.
M3.1 Flash has only 3 comments in the Product Hunt comment section.
25 more timeline events
All Facts (20)
M3.1 Flash can be called via the AI SDK, Chat Completions, Responses, and the Anthropic Messages API, with the model name minimax/minimax-m3.1-flash.
75%VerifiedThe official name of MiniMax M3.1 Flash is `MiniMax-M3.1-Flash`, positioned as a frontier multimodal coding model with a 1 million context window and adjustable thinking depth.
65%Partially VerifiedMiniMax M3.1 Flash is available on AIHubMix for token-billed calls under the model ID `minimax-m3.1-flash`, with input priced at $0.165 per million tokens and output at $0.66 per million tokens.
65%UnverifiedMiniMax M3.1 Flash uses a MoE architecture with approximately 428B total parameters and approximately 23B activated per token.
60%UnverifiedvLLM integrated three serving-stack configurations for M3.1 Flash: native NVFP4 KV cache, NVFP4 MoE weights, and DSpark speculative decoding.
50%UnverifiedEngineering improvements related to MiniMax M3.1 Flash are being upstreamed to vLLM.
50%UnverifiedWhether a model can be delivered by a host under an SLA is an important signal of its engineering maturity.
50%Unverified@inferact provides SLA-guaranteed inference hosting services for MiniMax M3.1 Flash.
50%UnverifiedM3.1 Flash has only 3 comments in the Product Hunt comment section.
50%UnverifiedThe officially stated 3x speed and 70% cost reduction are vendor-supplied figures, and actual performance depends on task type, context length and concurrency conditions.
50%UnverifiedM3.1 Flash reduces costs on agent tasks by 70%.
50%UnverifiedM3.1 Flash's online speed is about 3 times that of the previous generation.
50%UnverifiedM3.1 Flash received 112 votes and ranked 4th on Product Hunt.
50%UnverifiedM3.1 Flash's output speed reaches 200+ tokens per second.
50%UnverifiedMiniMax M3.1 Flash targets scenarios such as agentic tasks, coding, computer use, and cybersecurity.
50%UnverifiedMiniMax M3.1 Flash is currently offered only in API form, with weights still to be released later.
50%UnverifiedIf you are already using the MiniMax Flash series, M3.1 is worth an A/B test, focusing on latency and stability in long-text scenarios.
50%UnverifiedThe version change from M3 to M3.1 is a minor-version iteration, implying targeted tuning rather than an architecture-level rewrite.
50%UnverifiedIf you are still waiting to see, the information in this M3.1 Flash release alone is not enough to support a migration decision; you need to wait for third-party evaluations and real-world cases.
50%UnverifiedThe official positioning of M3.1 Flash is as an agentic model.
50%