Skip to content
Neural Daily

ModelsNews

MiniMax M3.1 Flash: cheap agent model with 1M context

China's MiniMax released M3.1 Flash on Monday. The company says it is a generational leap in agentic tasks, coding, computer use and cybersecurity. Online inference is nearly three times faster, and usage costs are 70% lower than its predecessor.

By: Mikola Máté Patrik · · 2 min read

MiniMax M3.1 Flash key visual: a bunny astronaut on a rocky surface in front of a red planet
Image: MiniMax · Press kit · Official launch image from the MiniMax blog · Source

What it can do

According to MiniMax's blog, M3.1 Flash scores 73.7% on the DeepSWE v1.1 coding benchmark and 76.2 on BankerToolBench (MiniMax). It scores 61.7 on OSWorld V2 and 29.6 on Agents' Last Exam (ALE), which MiniMax says is the best result among Chinese models on both. It reaches 82% on CyberGym, a vulnerability discovery test.

The API docs list a 1 million token context window with text, image and video input (MiniMax API Docs). Thinking depth is tunable with the effort parameter (low, medium, high, xhigh, max), with max as the default. The model is served on both OpenAI-compatible and Anthropic-compatible APIs as MiniMax-M3.1-Flash.

A fun detail: MiniMax says Space Bunny, a model that ran anonymously and processed more than 200 trillion tokens, was an early version of M3.1 Flash.

MiniMax benchmark table comparing M3.1 Flash with DeepSeek V4.1, Kimi-K3, GLM-5.3, GPT-5.6 Sol and Opus 5
MiniMax · Press kit · Official benchmark table from the launch post · Source

Source: MiniMax. The company's own benchmark table from the M3.1 Flash announcement, with notes on evaluation settings.

What it costs

Pricing is flat: $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. MiniMax says overall usage cost on agentic tasks is 70% lower than its predecessor. Output speed can exceed 200 tokens per second, with the median consistently above 170.

MiniMax chart of DeepSWE v1.1 score versus average cost per task for M3.1 Flash and rival models
MiniMax · Press kit · Official benchmark chart from the launch post · Source

Source: MiniMax. DeepSWE v1.1 score versus average cost per task, based on public DeepSWE leaderboard data as of September 22, 2026.

Where to get it

M3.1 Flash is available through the MiniMax API, the M Plan subscription and the MiniMax Code desktop app. Weights will be released later, as will the larger Pro model. Reuters also reported the launch (StreetInsider / Reuters).

Why it matters

These are MiniMax's own numbers and have not been independently verified. If the DeepSWE result holds up in practice, the $0.15 input price makes M3.1 Flash a strong candidate for background agents and high-volume coding work. The Anthropic-compatible endpoint makes it quick to try in existing agent harnesses.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles