MiMo-V2-Flash

MiMo-V2-Flash is a Mixture-of-Experts language model from Xiaomi with 309B total parameters and 15B active parameters, supporting a 262K token context window.

Strengths

The sparse MoE architecture keeps computational requirements low while leveraging a large parameter pool, and the extended context window enables processing of long documents in a single pass.

Best for

Applications requiring long-context processing and cost-efficient inference where sparse activation patterns are acceptable.

Limitations

As an open-source foundation model, it may require additional fine-tuning for specialized tasks, and sparse MoE models can have less predictable performance across all workload types compared to dense alternatives.