Nake
Settings
Language
ThemeAtelier
Privacy policyTerms of serviceRefund policy
Premium-50%50% off
Lite

Premium membership from $19 per month

Premium benefitsMore benefits
Fast generationHD outputAll image toolsAll video tools
Nake
  • Home
  • For You
  • AI Video
  • AI Image
  • Anime / Manga
  • Video Edit
  • Video Extend
  • AI Tools
  • Generate
  • Agent
  • Explore
  • Home
  • For You
  • AI Tools

Create & chat

  • AI Video
  • AI Image
  • Anime / Manga
  • AgentBeta
  • Explore
  • Generate
  • Video Edit
  • Video Extend+5s
Upgrade for more
    • Home
    • For You
    • Create
    • Pricing
    • More

    minimax m3 1 flash

    minimax m3 1 flash is MiniMax's next-generation flash model, delivering a generational leap in agentic tasks, coding, computer use, and cybersecurity. It processes over 200 trillion tokens, offers online inference nearly three times faster, and reduces usage costs by 70% compared to its predecessor. The model is now available through the MiniMax API, M Plan, and MiniMax Code, with model weights to be released later.

    Official documentation
    minimax logo
    Publisher identity: minimax logo.

    On this page

    minimax m3 1 flash: Model Overview

    minimax m3 1 flash represents a significant advancement in AI model efficiency and capability, designed specifically for complex agentic tasks and real-world automation.

    Next-Generation Flash Architecture

    minimax m3 1 flash is built on a next-generation flash architecture that processes more than 200 trillion tokens. The model, codenamed Space Bunny 🐰, operated anonymously during its early development phase. This architecture delivers a generational leap in agentic tasks, coding, Computer Use, and cybersecurity, while making online inference nearly three times faster and reducing usage costs by 70%. The model achieves output speeds of 200+ tokens per second in everyday use, with median (P50) speeds consistently reaching 170+ TPS through more aggressive sparsity optimization. This efficiency makes it suitable for production environments where both speed and cost are critical factors.

    Benchmark Performance Across Domains

    minimax m3 1 flash demonstrates strong performance across multiple benchmarks. On AutomationBench-AA, it ranks among top performers for complex agentic tasks, completing real enterprise workflows across SaaS systems in sales, marketing, operations, customer service, finance, and HR. It scores 76.2 on BankerToolBench, excelling at multi-step planning, document consistency, financial compliance, and chained tool calls. For coding, it achieves 73.7% on DeepSWE v1.1, placing it alongside top models for complex software tasks requiring independent code understanding, feature development, issue fixing, and validation through testing. In computer use, it scores 61.7 points on OSWorld V2 and 29.6 points on Agents' Last Exam (ALE), ranking first among Chinese models on both benchmarks.

    minimax m3 1 flash: Capabilities and Access

    minimax m3 1 flash offers a range of capabilities for agentic tasks, coding, and computer use, with multiple access options available.

    Agentic Task Execution and Automation

    This model excels in complex agentic tasks and automation. It can complete real enterprise workflows across SaaS systems, solving practical problems in sales, marketing, operations, customer service, finance, and HR from end to end. The model demonstrates strong capabilities in interacting with real environments, using tools, and executing long sequences of tasks—moving beyond understanding a task to taking action and delivering results. In cybersecurity, it achieves a strong score of 82% on CyberGym, demonstrating leading capabilities in vulnerability discovery and validation. Internal security experts found the model performed strongly in code auditing, vulnerability patching, penetration testing, malware analysis, and reverse engineering.

    Access Options and Pricing

    This model is now available through the MiniMax API, M Plan, and MiniMax Code. The model offers flat pricing per 1M tokens: $0.15 for input, $0.60 for output, and $0.015 for cached input. For agentic tasks, the model's overall usage cost is 70% lower than its predecessor. Model weights will be released later. Users can access the model through the MiniMax API at platform.minimax.io, subscribe to M Plan at minimax.io/m-plan, or download MiniMax Code for desktop at agent.minimax.io/download. The model supports a 1 million token context window and tunable thinking depth that can be adjusted on the fly from low to max.

    Availability

    This model is currently available through multiple access points, with additional availability planned for the future.

    Current Access Methods

    This model is now available through the MiniMax API, M Plan, and MiniMax Code. The model can be accessed via the MiniMax API at platform.minimax.io, through subscription to M Plan at minimax.io/m-plan, or by downloading MiniMax Code for desktop at agent.minimax.io/download. MiniMax Code is a command-line coding agent that runs models like M3.1 Flash directly from a terminal session. Setup involves installing the binary, sourcing it into your shell profile, authenticating with an API key tied to a MiniMax coding plan, and launching it with a single command. The model appears as a selectable option in the model picker.

    Future Availability and Open Weights

    MiniMax has indicated plans to release this model as open weights on Hugging Face, though no firm date has been confirmed. The model is currently in preview and only accessible through MiniMax's own coding plan, used via the M-Code terminal agent and API. MiniMax has also announced that Pro, which pushes the boundaries of intelligence, will arrive later. This suggests a tiered approach to model availability, with the Flash variant focusing on the balance between efficiency and capability, while Pro will target maximum intelligence. Users should monitor MiniMax's official channels for updates on open weights release and additional access methods.

    Configurations

    This model offers flexible configuration options for different use cases and performance requirements.

    Thinking Depth Configuration

    This model offers tunable thinking depth ranging from low to max, which can be changed on the fly within a session rather than being fixed at the start. This allows developers to trade speed for reasoning depth mid-session without committing to one mode for an entire task. The low thinking depth configuration prioritizes speed and efficiency for simpler tasks, while the max thinking depth configuration dedicates more computational resources to complex reasoning. This flexibility makes the model suitable for a wide range of applications, from quick code completions to deep analytical tasks requiring extensive reasoning.

    Context Window and Token Handling

    This model supports a 1 million token context window, large enough to handle long files, extended conversations, or large codebases in a single session. The model processes more than 200 trillion tokens overall, with output speeds reaching 200+ tokens per second in everyday use and median (P50) speeds consistently reaching 170+ TPS. This is achieved through more aggressive sparsity optimization. The model offers flat pricing per 1M tokens: $0.15 for input, $0.60 for output, and $0.015 for cached input. For agentic tasks, the model's overall usage cost is 70% lower than its predecessor, making it cost-effective for large-scale deployments.

    Practical Notes

    Practical considerations for using this model in real-world applications, based on hands-on testing and benchmark results.

    Performance on Analytical vs. Creative Tasks

    Hands-on testing of this model revealed strong performance on analytical tasks like spotting subtle SQL bugs and reasoning through complex decision scenarios with structured JSON output. In a drone decision test, the model correctly read a low battery level, identified a landing hazard, and chose a safe landing pad, though it missed the exact position of a thin power line. In a SQL debugging test on a six-join legacy DB2 query, the model found a boundary-date bug almost immediately. On a larger generative task (building a full 3D game), it succeeded on core mechanics but fell short on polish, specifically camera positioning, suggesting stronger performance on precision reasoning than on holistic creative execution.

    M-Code Agent Features and Workflow

    The M-Code terminal agent, which runs this model, stood out for speed and visibility into its own work. It gives users more control compared to other terminal coding agents, including mid-task pausing. Tasks that might take competing agents a long while to loop through were resolved quickly, and the system's record of what it had already done was easy to inspect without digging through logs. The agent supports requesting permission to execute actions in bulk and works through multi-step tasks like file creation, verification, and self-checking before handing the result back. This workflow makes it suitable for complex development tasks where oversight and control are important.

    FAQ

    Frequently asked questions about this model.

    What should I know about minimax m3 1 flash?

    minimax m3 1 flash is MiniMax's next-generation flash model that delivers a generational leap in agentic tasks, coding, computer use, and cybersecurity. It processes over 200 trillion tokens, offers online inference nearly three times faster, and reduces usage costs by 70% compared to its predecessor. The model is now available through the MiniMax API, M Plan, and MiniMax Code, with model weights to be released later. It supports a 1 million token context window and tunable thinking depth that can be adjusted on the fly. The model achieves strong benchmark scores across multiple domains, including 73.7% on DeepSWE v1.1 for coding, 61.7 points on OSWorld V2 for computer use, and 82% on CyberGym for cybersecurity.

    Original sources

    • MiniMax M3.1 Flash: Redefining Agent Performance and Cost - MiniMax Research | MiniMax
    • MiniMax M3.1 Flash Preview: Hands-On Tests With M-Code Agent | MindStudio

    Related reading

    AI ToolsText to ImageImage to Video

    Read the sources for minimax m3 1 flash

    Creative tools & scenes
    • minimax h3 nsfw
    • midjourney nsfw
    • ltx 2.3 nsfw
    • minimax nsfw
    • mistral nsfw
    • mochi 1 nsfw