# MiniMax Launches M3.1 Flash Model for Enhanced Agentic Performance The new M3.1 Flash model delivers triple the inference speed and 70% lower costs. - **Source:** [Hacker News](https://news.ycombinator.com) - **Published:** 2026-10-19 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/184b1f4a-7d53-4f57-962f-95c428c0e94b ## Launch of M3.1 Flash MiniMax has officially released M3.1 Flash, a next-generation model designed to enhance agentic task execution. This release follows the experimental phase of Space Bunny, an early version that processed over 200 trillion tokens while operating anonymously. The model is now accessible through the MiniMax API, M Plan, and the MiniMax Code platform. ## Performance and Speed Metrics The M3.1 Flash model achieves significant speed improvements, with output speeds reaching over 200 tokens per second. Median performance consistently hits 170 tokens per second during everyday operations. These gains are driven by aggressive sparsity techniques, which allow the model to maintain high efficiency while handling complex, real-world computational and agentic workloads. ## Cost Efficiency Improvements Operational costs for agentic tasks have been reduced by 70% compared to the predecessor. The current 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 competitive economic model for developers integrating these capabilities into their own applications. ## Benchmarking Against Industry Standards M3.1 Flash demonstrates performance parity with leading models across three specific benchmarks. These include Automation-Bench for complex task execution, DeepSWE for software development, and Agents’ Last Exam for computer use. By achieving these results at a lower cost, the model establishes a new standard for balancing high-level intelligence with operational affordability. ## Agentic and Coding Capabilities The model is specifically optimized for agentic tasks, coding, and cybersecurity applications. It excels in computer use scenarios, enabling it to perform complex sequences that require real-world interaction. These capabilities allow the model to function effectively in environments where software development and automated task execution are critical components of the workflow. ## Future Roadmap and Releases While M3.1 Flash is currently available via API, the company plans to release model weights at a later date. Additionally, the Pro version of the model is currently under development. This upcoming iteration aims to push the boundaries of intelligence further, complementing the efficiency-focused design of the Flash series currently in production. ## Technical Infrastructure Evolution The transition from the Space Bunny prototype to the M3.1 Flash production model highlights the company's focus on scaling inference efficiency. By processing massive token volumes during the testing phase, the developers refined the sparsity mechanisms. This infrastructure evolution ensures that the model remains performant even as it scales to handle diverse, high-value tasks. ## Strategic Market Positioning By prioritizing the balance between capability and cost, MiniMax positions M3.1 Flash as a viable tool for developers requiring high-speed inference. The focus on agentic tasks suggests a strategic shift toward models that can execute complex workflows autonomously. This approach aims to lower the barrier for integrating advanced AI into commercial software development. ## Sources covering this story - https://www.minimax.io/blog/minimax-m31f — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._