Xiaomi Unveils MiMo-V2-Pro, a Powerful AI Model Rivaling GPT-5.2 at a Fraction of the Cost

Xiaomi, known for smartphones and EVs, has unveiled MiMo-V2-Pro, a 1-trillion parameter AI model that matches the performance of GPT-5.2 and Claude Opus 4.6 at a significantly lower cost. The model's sparse architecture and advanced capabilities make it well-suited for autonomous digital workflows.

Details

Xiaomi's MiMo-V2-Pro is a groundbreaking AI model that challenges the dominance of OpenAI and Anthropic in the large language model (LLM) space. The model's sparse architecture, with only 42 billion active parameters during any single forward pass, allows it to maintain a massive 1-million token context window while keeping costs low. Its 7:1 hybrid attention mechanism enables efficient processing of large amounts of data, acting like an 'expert researcher' by focusing on the most critical 15% of the information. Additionally, the model's multi-token prediction (MTP) capability dramatically reduces the latency required for autonomous digital workflows. Xiaomi is positioning MiMo-V2-Pro as the 'brain' for digital agents, terminals, and APIs, with third-party testing showing exceptional performance in agentic environments and a low hallucination rate. This model has the potential to reshape how developers build autonomous AI systems, offering a streamlined approach that eliminates the need for complex chunking logic.