MiMo-V2.5 - Xiaomi's all-modal agent large model series

What is MiMo-V2.5?

MiMo-V2.5 is a new generation of multimodal agent models launched by Xiaomi's large model team, including four versions: V2.5, V2.5-Pro, V2.5-TTS, and V2.5-ASR. MiMo-V2.5 is a native multimodal agent, supporting image, audio, and video understanding and action, with a context size of up to 1MB. MiMo-V2.5-Pro is designed for long-term complex tasks, capable of stably completing nearly a thousand rounds of tool calls. In terms of software engineering capabilities, it rivals Claude Opus 4.6 and GPT-5.4, with a 42%~50% improvement in token efficiency. The entire MiMo-V2.5 model series will soon be open source.

Main functions of MiMo-V2.5

• MiMo-V2.5

  • Native Full-Modal AgentIt supports simultaneous viewing, listening, and reading, transforming understanding into action and covering multimodal perception of images, audio, and video.
  • General Agent ScenarioIt can handle simple to moderately complex daily tasks, and the API cost is reduced by about 50% compared to the previous generation.
  • Multimodal perception upgradeIts cross-modal reasoning, video understanding, and graph analysis capabilities surpass those of its predecessor, MiMo-V2-Omni, and it approaches top-tier closed-source models in benchmarks such as VideoMME, CharXiv, and MMMU-Pro.

  MiMo-V2.5-Pro

  • Long-term complex tasksIt can reliably complete nearly a thousand rounds of tool calls in a single professional-level task, with significantly enhanced command compliance and understanding of fuzzy commands.
  • Complex Software EngineeringIt supports large-scale codebase development, multi-language programming, code debugging and performance optimization, and can independently deliver projects such as compilers and video editors.
  • Autonomous Project DeliveryImplement a complete SysY compiler from scratch in Rust (4.3 hours, 672 calls, perfect score of 233/233 on the hidden test set); Build a working multitrack video editor web application with only simple instructions (8,192 lines of code, 1,868 calls, completed in 11.5 hours).

  Voice capabilities

  • V2.5-TTS SeriesThe naturalness of speech synthesis and support for multiple languages/dialects/voices have been fully upgraded.
  • V2.5-ASRImproved speech recognition accuracy and real-time performance, supporting multiple languages and dialects.

How to use MiMo-V2.5

• Web-based experienceVisit the MiMo Studio official website, register and log in, then select the MiMo-V2.5 model to start the conversation.
• API Development and IntegrationVisit the API Open Platform https://platform.xiaomimimo.com/#/token-plan to register a developer account, create an application, and obtain an API Key.

Key information and usage requirements of MiMo-V2.5

• Access methods
  • MiMo Studio Experience: https://aistudio.xiaomimimo.com
  • API calls: https://platform.xiaomimimo.com
• Open source projectMiMo-V2.5-Pro and MiMo-V2.5 will soon be open source globally.
• Pricing and Billing
  • MiMo-V2.5: 1x Credits (1 Token = 1 Credit)
  • MiMo-V2.5-Pro: 2x Credits (1 Token = 2 Credits)
• Context windowThe entire series supports ultra-long contexts of 1M Tokens (approximately 750,000 words), and long document analysis will no longer incur additional charges.

The core advantages of MiMo-V2.5

• Token Efficiency RevolutionWhen achieving the same score on the ClawEval benchmark, V2.5-Pro saves 42% of tokens compared to Kimi K2.6, and V2.5 saves 50% of tokens compared to Muse Spark, significantly reducing the cost of large-scale deployments.
• Long-term task stabilityWhen paired with a suitable runtime framework, it can stably complete long-term tasks involving nearly a thousand rounds of tool calls in a single session, and maintain logical consistency and self-correction over an extremely long period (such as self-diagnosis and recovery after the 512th round of refactoring in a compiler task).
• Full-modal native fusionText, image, audio, and video capabilities are integrated into a single model, eliminating the need to switch between multimodal and text models.
• Reasoning speedV2.5 has an average inference speed of 100~150 tokens/s, while V2.5-Pro has a speed of 60~80 tokens/s. The basic version is more suitable for latency-sensitive scenarios.

MiMo-V2.5 project address

• Project official website: https://platform.xiaomimimo.com/docs/news/v2.5-news

Comparison of MiMo-V2.5 with similar competing products

Application scenarios of MiMo-V2.5

• Complex software developmentIt can independently build and debug large-scale engineering projects such as compilers, video editors, and web applications from scratch, replacing the workload of human experts for several days to several weeks.
• Multimodal content analysisUpload photos of your refrigerator to get recipe suggestions, analyze video tutorials to generate step summaries, and record meetings to extract to-do items.
• Scientific and Engineering AutomationAnalog circuit EDA design, performance optimization, long-range analysis of academic literature, and cross-document reasoning.
• Intelligent Customer Service and Interaction: In-vehicle, smart home, and real-time interactive systems based on full-modal perception (voice + image + text).
• Enterprise-level Agent DeploymentIn large-scale API call scenarios, high token efficiency enables cost-controllable automated workflows.