Xiaomi's AI Stuns World: Open Model Tops Proprietary Giants

Xiaomi's new MiMo-V2.6-Pro has claimed the top spot as the world's leading open-weight language model, according to Artificial Analysis. This achievement places it ahead of proprietary models such as xAI's Grok 4.6 and Google's Gemini 3.8 Flash. Notably, MiMo-V2.6-Pro is open-weight and MIT-licensed, allowing free download from Hugging Face for customization and self-hosting.

The model also offers competitive API pricing, with input tokens at $0.435 per million and output tokens at $0.87 per million. Accompanying it are MiMo-V2.6-Flash, a smaller, more affordable option, and MiMo-V2.6-Pro-UltraSpeed for enhanced performance. Xiaomi's rapid AI advancement includes prior releases like MiMo-V2.5-Pro in April 2026 and MiMo Code in June 2026.

A significant technical development behind V2.6 is extensive reinforcement learning (RL), costing millions, to enhance agent capabilities. Xiaomi employed a "You Only RL Once" strategy, mixing various domains and harnesses in single training runs. This approach, using asynchronous GRPO, aims to optimize agent performance by rewarding clean solutions over mere task completion.

While MiMo-V2.6-Pro leads open-weight models, it does not outperform all proprietary systems, with models like Anthropic's Claude Opus 5 and OpenAI's GPT-5.6 Sol still excelling in specific benchmarks. However, Xiaomi's accessible pricing and open-weight strategy present a compelling alternative for enterprises seeking capable AI solutions.