AI News Feed
Market watch
Professor Calls for Abolishing Business Schools, Saying They Favor Profit Over Public LifeTechRadar Releases Oct. 20 Hints and Answers for Quordle, Strands and ConnectionsCramer Says AI Data-Center Rebound Is Not Yet a Buy SignalAI Tool Predicts Cancer Survival From Single Selfie, Study SaysTechRadar: France's Foreign Legion May Get Robot Mules From Retired TanksHugging Face policy chief says embedded AI evaluators are no easy fix
Professor Calls for Abolishing Business Schools, Saying They Favor Profit Over Public LifeTechRadar Releases Oct. 20 Hints and Answers for Quordle, Strands and ConnectionsCramer Says AI Data-Center Rebound Is Not Yet a Buy SignalAI Tool Predicts Cancer Survival From Single Selfie, Study SaysTechRadar: France's Foreign Legion May Get Robot Mules From Retired TanksHugging Face policy chief says embedded AI evaluators are no easy fix
Large Language Models

MiniMax releases M3.1 Flash, claims 70% lower AI agent costs

MiniMax released M3.1 Flash, claiming 70% lower agent-task costs and nearly triple the inference speed of M3.

AI News Feed · 2026-10-20 08:10

The model is aimed at AI agents, which often need dozens of model calls to complete a single job. MiniMax said Flash reflects its effort to balance efficiency and capability. The release lists 428 billion parameters, barely changed from M3, and credits more aggressive sparsity at inference time for the speed increase. Reasoning remains on for every request, with developers choosing one of five effort levels from Low to Max. M3.1 Flash accepts text, image and video input, and prompts can run to 1 million tokens.

In MiniMax’s own testing, M3.1 Flash scored 61.7 on OSWorld V2, which measures how well a model can operate a computer; MiniMax said no Chinese model has done better. It scored 76.2 on BankerToolBench, a banking workflow test.

For security work, MiniMax pointed to an 82% score on CyberGym, a benchmark built by University of California, Berkeley researchers to test whether AI agents can reproduce real software vulnerabilities. The company listed code auditing, penetration testing and malware analysis among jobs the model can take on.

At a flat 15 cents per million input tokens, M3.1 Flash lines up directly against DeepSeek’s V4.1-Flash, which charges the same rate off-peak. DeepSeek said over the weekend that its updated V4.1-Flash resolved 77% of tasks on the DeepSWE v1.1 software engineering test. M3.1 Flash managed 73.7%. Neither result has been checked independently.

MiniMax has traded in Hong Kong since a $619 million initial public offering in January. Bloomberg reported that its shares fell after the original V4.1-Flash launched in September.

The launch also settles the identity behind Space Bunny, an anonymous model that appeared on OpenRouter Inc.’s model routing service in late September. Community testers linked it to MiniMax by probing its tokenizer, and MiniMax confirmed that Space Bunny was an early M3.1 Flash build. More than 200 trillion tokens went through it while its maker stayed unnamed, according to the launch post.

M3.1 Flash is available now through the MiniMax API, its M Plan subscription and its MiniMax Code desktop app. Output costs 60 cents per million tokens and cached input 1.5 cents. MiniMax said it will release the Flash weights later and that a more capable Pro model is also coming.