MiMo-V2.6-Flash is an open-source Mixture-of-Experts model from Xiaomi with 309B total parameters and 15B activated per token. Its hybrid attention (sliding-window layers interleaved with global attention) keeps long-context serving efficient across a 1M-token context window. It accepts text and image input and is suited for coding, agent workflows and high-volume tool use. Reasoning is supported and enabled by default.