MiniMax Launches M3.1 Flash: Redefining Agent Performance and Cost Efficiency

Summary:MiniMax has officially launched the next-generation agent model, M3.1 Flash, which delivers significant performance improvements in complex agent tasks, coding, computer use, and cybersecurity. Compared to its predecessor, M3.1 Flash achieves nearly triple the online inference speed and reduces usage costs by 70%. The model excels in benchmarks such as AutomationBench-AA, DeepSWE v1.1, and CyberGym, demonstrating strong capabilities in enterprise workflows, software development, and vulnerabilit

---

Technical Breakthroughs and Core Features

MiniMax M3.1 Flash is the company's next-generation agent model, designed to redefine performance and cost-efficiency in complex tasks. The key technical breakthroughs include:

1. Complex Task Handling and Automation

   • M3.1 Flash excels in the AutomationBench-AA benchmark, capable of handling real enterprise workflows across SaaS systems, effectively solving practical problems in sales, marketing, operations, customer service, finance, and HR.
   • It scores 76.2 on BankerToolBench, demonstrating strong capabilities in multi-step planning, document consistency maintenance, financial compliance, and chained task calls.

2. Coding and Software Engineering

   • In the DeepSWE v1.1 test, M3.1 Flash achieves an accuracy of 73.7%, placing it alongside top models. This test focuses on complex software tasks that require models to independently understand code, develop features, fix issues, and validate their work through testing.
   • M3.1 Flash demonstrates strong execution across long sequences of steps, supporting complex software development workflows at a lower cost.

3. Computer Use

   • It scores 61.7 on OSWorld V2, which measures AI computer operation and cross-application task execution.
   • It scores 29.6 on Agents’ Last Exam (ALE), evaluating performance on real professional work and complex tasks. M3.1 Flash ranks first among Chinese models on both benchmarks, showcasing its strong capabilities in interacting with real environments, using tools, and executing long sequences of tasks.

4. Cybersecurity

   • It achieves a strong score of 82% on CyberGym, demonstrating leading capabilities in vulnerability discovery and validation. In-depth evaluations by internal security experts show that the model performs well in code auditing, vulnerability patching, penetration testing, malware analysis, and reverse engineering, enabling efficient, low-cost support for building complex security systems.

Engineering Trade-offs and Real-world Performance

M3.1 Flash optimizes inference efficiency through more aggressive sparsity. In everyday use, output speeds can reach 200+ TPS (tokens per second), with median (P50) speeds consistently reaching 170+ TPS.

In terms of pricing, M3.1 Flash's overall usage cost for agentic tasks is 70% lower than its predecessor. The flat pricing per 1M tokens is $0.15 for input, $0.60 for output, and $0.015 for cached input.

Developer Takeaways and Deployment Recommendations

M3.1 Flash is now available through the MiniMax API, M Plan, and MiniMax Code. The model weights will be released later. Developers can download the MiniMax Code for desktop via https://agent.minimax.io/download or subscribe to M Plan via https://www.minimax.io/m-plan. API access is available at platform.minimax.io.

Conclusion

M3.1 Flash represents MiniMax's ongoing exploration of the balance between efficiency and capability. The Pro version, which pushes the boundaries of intelligence, will also be released later.

---

Source: MiniMax Blog (2026-10-19)

— END —

Tags: #MiniMax Follow #Intelligent Agent Follow #AI Reasoning Follow #Cost Optimization Follow #Cybersecurity Follow