Xiaomi Releases Three MiMo V2.6 Models on AI Gateway

Xiaomi has released three models from the MiMo V2.6 series on AI Gateway. These models combine coding, reasoning, and tool use, and natively understand text, images, audio, and video. They support a 1M token context window, making them suitable for long codebases or multi-session agent tasks, and can generate up to 128K tokens of output.

Model Features

You can choose one of three models depending on the workload.

• xiaomi/mimo-v2.6-pro: A large sparse Mixture-of-Experts (MoE) model that activates 42B parameters per token out of 1.02T total parameters, optimized for complex software engineering and long-running agent tasks.
• xiaomi/mimo-v2.6-flash: An efficient model that activates 15B parameters per token out of 309B total parameters, suitable for multimodal automation and everyday agent workflows.
• xiaomi/mimo-v2.6-pro-ultraspeed: Delivers the same performance as the Pro model while increasing output speed by up to 20x, used for interactive and latency-sensitive workflows.

Usage and Platform Features

To use in coding agents, install the latest Vercel CLI and select the model. AI Gateway provides a unified API for model calls, supporting usage and cost tracking, retry and failover configuration, high-availability setup, and more. It also includes features such as custom reporting, API key budget management, and routing rules.