ARTIFICE DAILY AI + CYBER SECURITY BRIEFING

Cybersecurity

MiniMax launches M3.1 Flash with weights to follow

MiniMax said it had released M3.1 Flash, an agentic model it claimed improved speed, cost and benchmark results across coding, computer use and cybersecurity, with weights due later.

MiniMax

MiniMax said it had rolled out M3.1 Flash and called Space Bunny 🐰 an earlier version of the model. The company said Space Bunny had handled over 200 trillion tokens while operating anonymously.

MiniMax said M3.1 Flash was its next-generation Flash model for agentic work, coding, computer use and cybersecurity. It said the model made online inference nearly three times faster and cut usage costs by 70%.

The company listed flat prices of $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. It said everyday output speeds could reach 200+ TPS, with median speeds of 170+ TPS.

MiniMax said the model scored 76.2 on AutomationBench-AA, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam and 82% on CyberGym. It also said internal security staff had evaluated code auditing, patching, penetration testing, malware analysis and reverse engineering.

The model was made available through the MiniMax API, M Plan and MiniMax Code, and the company said the weights would be released later. It added that Pro would arrive later too.

Named in this story

Companies

MiniMax
said it had rolled out M3.1 Flash and set pricing and access terms

Products and systems

M3.1 Flash
the model MiniMax announced and benchmarked
Space Bunny 🐰
an earlier version of the model that had handled over 200 trillion tokens
MiniMax API
one channel through which the model was made available
M Plan
one subscription route for access
MiniMax Code
the desktop product through which the model was offered

How the source tells it

The article reads as a promotional launch note, with MiniMax’s own benchmark and pricing claims doing most of the work.

  • hype loaded superlatives and a claimed leap in capability across several benchmarks
  • vendor boosterism the company’s own performance, pricing and availability claims are presented without outside corroboration