MiniMax has officially launched M3.1 Flash, a new AI model designed for agentic tasks and automation. According to the company, M3.1 Flash delivers a “generational leap” in performance across automation, coding, computer use, and cybersecurity while making online inference nearly three times faster and reducing usage costs by 70% compared to its predecessor.
The model is built on an early version called Space Bunny, which processed more than 200 trillion tokens while operating anonymously.
Performance Benchmarks
M3.1 Flash demonstrates competitive performance across multiple evaluation frameworks. On AutomationBench-AA, which tests real enterprise workflows across SaaS systems, the model ranks among top performers. It can complete end-to-end tasks in sales, marketing, operations, customer service, finance, and HR.
For software development, M3.1 Flash achieves 73.7% on DeepSWE v1.1, a benchmark focused on complex coding tasks requiring models to understand code, develop features, fix issues, and validate work through testing. The model scores 76.2 on BankerToolBench, excelling at multi-step planning, document consistency, financial compliance, and chained tool calls.
On computer use benchmarks, M3.1 Flash scores 61.7 points on OSWorld V2, which measures cross-application task execution, and 29.6 on Agents’ Last Exam (ALE), evaluating performance on real professional work. According to MiniMax, the model ranks first among Chinese models on both benchmarks.
In cybersecurity, M3.1 Flash achieves 82% on CyberGym, demonstrating capabilities in vulnerability discovery and validation. Internal security evaluations showed strong performance in code auditing, vulnerability patching, penetration testing, malware analysis, and reverse engineering.
Speed and Pricing
MiniMax has optimized inference efficiency through aggressive sparsity optimization. Output speeds reach 200+ tokens per second (TPS), with median speeds consistently at 170+ TPS.
Pricing is structured at $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cached input tokens—representing a 70% overall cost reduction for agentic tasks compared to the previous model.
M3.1 Flash is available through the MiniMax API, M Plan subscription, and MiniMax Code. Model weights will be released later. The company indicates that a Pro model pushing “the boundaries of intelligence” will arrive in the future.
Key facts
- M3.1 Flash reduces usage costs by 70% and makes inference nearly 3x faster than its predecessor
- The model scores 73.7% on DeepSWE v1.1 for software development and 82% on CyberGym for cybersecurity
- Output speeds reach 200+ tokens per second, with median speeds at 170+ TPS
- Flat pricing: $0.15 input, $0.60 output, $0.015 cached input per 1M tokens
- M3.1 Flash ranks first among Chinese models on OSWorld V2 and Agents’ Last Exam benchmarks