MiniMax Launches M3.1 Flash Model for Enhanced Agentic Performance
New model delivers triple the inference speed and 70% lower costs for developers.
1Launch of M3.1 Flash
MiniMax has officially released M3.1 Flash, a next-generation model designed for high-performance agentic tasks. This release follows the secret development phase of the Space Bunny project, which processed over 200 trillion tokens anonymously. The model is now accessible to users through the MiniMax API, M Plan, and the MiniMax Code platform.
2Efficiency and Speed Metrics
The M3.1 Flash model achieves nearly three times faster online inference compared to its predecessor. Through aggressive sparsity optimizations, the model reaches output speeds exceeding 200 tokens per second. Median performance consistently maintains over 170 tokens per second, ensuring high responsiveness for developers integrating the model into complex, real-time software applications.
3Significant Cost Reductions
Operational costs for agentic tasks have been reduced by 70% with this new release. The pricing structure is set at $0.15 per million input tokens and $0.60 per million output tokens. Cached input tokens are priced at $0.015, providing a highly competitive economic model for businesses scaling their AI-driven automation workflows.
4Benchmark Performance Standards
M3.1 Flash demonstrates parity with leading industry models across three critical benchmarks. These include Automation-Bench for complex agentic execution, DeepSWE for real-world coding tasks, and Agents’ Last Exam for computer use. By matching top-tier performance at a lower cost, the model establishes a new standard for balancing intelligence with operational efficiency.
5Agentic and Coding Capabilities
The model is specifically engineered to excel in agentic tasks, software development, and cybersecurity operations. It facilitates advanced computer use, allowing for the execution of complex workflows that require high-level reasoning. These capabilities position M3.1 Flash as a robust tool for developers building automated systems that interact with real-world digital environments.
6Strategic Product Roadmap
MiniMax continues to explore the balance between capability and efficiency through its Flash series. While M3.1 Flash focuses on speed and cost-effectiveness, the company has confirmed that a Pro version is currently in development. This future iteration will aim to push the boundaries of intelligence further, though specific release timelines remain unannounced.
7Deployment and Availability
Developers can currently integrate M3.1 Flash via the MiniMax API, M Plan, and MiniMax Code services. While the API is fully operational for immediate use, the company has indicated that the model weights will be released at a later date. This phased approach allows for immediate platform adoption while managing future open-access requirements.
8Impact on AI Development
The introduction of M3.1 Flash signals a shift toward prioritizing inference efficiency in agentic AI. By lowering the barrier to entry for complex task execution, MiniMax enables broader adoption of autonomous agents. The focus on high-speed, low-cost processing directly addresses the primary economic challenges currently facing organizations scaling their generative AI infrastructure.