Xiaomi Deprecates MiMo-V2.5-Pro Model, Upgrading API to V2.6-Pro with Multimodal Support
Xiaomi has officially deprecated its older MiMo-V2.5-Pro and MiMo-V2.5 models, automatically routing their API requests to the newer V2.6-Pro and V2.6-Flash versions. The upgrade to V2.6-Pro introduces native multimodal understanding capabilities, replacing the text-only input of the predecessor. This model lifecycle update simplifies Xiaomi's API ecosystem while granting developers access to more capable and cost-effective native multimodal models. It aligns with the broader industry trend of consolidating text and vision capabilities into unified foundational models. API requests for the deprecated models are automatically rerouted, and certain parameters like temperature and top_p are fixed in thinking mode for the new models. Additionally, the default max_completion_tokens for the flash variant has been significantly increased from 32,768 to 131,072.
## BACKGROUND
Multimodal AI models are artificial intelligence systems designed to process and understand multiple types of data modalities, such as text, images, and audio, simultaneously. API routing allows platforms to seamlessly transition developer traffic from deprecated software versions to newer, optimized iterations without requiring immediate code rewrites.