Overview
- Xiaomi launched MiMo‑V2.5 and MiMo‑V2.5‑Pro, which fold vision, speech, and video into one model family, with access live through the MiMo API and limited availability in AI Studio.
- The company reports near top‑tier results on coding and agent tasks, including a 57.2% score on SWE‑bench Pro, a test where models fix real software bugs, though it still trails on the hardest reasoning exams.
- Both models support a 1M‑token context window with no extra fee, with MiMo‑V2.5‑Pro priced at $1.00 per million input tokens and $3.00 per million output tokens, and the base model at $0.40 input and $2.00 output.
- Xiaomi says MiMo‑V2.5‑Pro uses about 42% fewer tokens than Kimi K2.6 at similar scores and the base model uses nearly half the tokens of Muse Spark, which can lower bills for large, long‑running workflows.
- Xiaomi highlights a rapid release cadence backed by a $8.7 billion AI commitment, growing usage on OpenRouter after a Hermes free‑access push, and plans to open‑source the models as it trains the next generation.