MiniMax Launches M3.1 Flash: an Agent Model Three Times Faster and 70% Cheaper

MiniMax has officially released its M3.1 Flash language model, pitched as a next-generation "Flash" model built for agentic workloads. According to the company, an early version operated anonymously under the name Space Bunny 🐰 and processed more than 200 trillion tokens along the way.

What's improved

MiniMax claims M3.1 Flash makes a "generational leap" across several fronts at once: complex agentic tasks, coding, Computer Use and cybersecurity. At the same time, online inference is said to be nearly three times faster, with usage costs down by 70%.

In business-process automation the model ranks among the leaders on AutomationBench-AA, the company says, handling real enterprise workflows across SaaS systems — from sales and marketing to finance and HR. On BankerToolBench it scored 76.2, coping with multi-step planning, financial compliance and chained tool calls.

Code, computer and security

For software engineering, the model reached 73.7% on DeepSWE v1.1, a benchmark that requires understanding code, building features, fixing issues and validating work through tests. In computer use it scored 61.7 on OSWorld V2 and 29.6 on Agents' Last Exam, taking first place among Chinese models on both, according to MiniMax.

Cybersecurity is singled out: 82% on CyberGym, while the firm's in-house experts reported strong results in code auditing, vulnerability patching, penetration testing, malware analysis and reverse engineering.

Speed and pricing

Through more aggressive sparsity, output speeds reach 200+ tokens per second, with the median holding above 170. Pricing is flat: $0.15 per million input tokens, $0.60 per million output tokens and $0.015 for cached input.

The model is already available via the MiniMax API, the M Plan subscription and the MiniMax Code environment; model weights are promised later. The company also teased a future Pro model that will "push the boundaries of intelligence."

Source: Hacker News