MiMo V2 Flash Free

MiMo V2 Flash Free is the OpenCode Zen-hosted variant of Xiaomi's open-weights MiMo-V2-Flash line, designed to remove per-token pricing as a constraint for long-context text work. It is built on a Mixture-of-Experts architecture with roughly 309 billion total parameters, of which about 15 billion are active per token, and the weights are distributed under an MIT license. Throughput is the headline optimization: the underlying model uses Multi-Token Prediction together with self-speculative decoding, and the release page advertises sustained generation in the neighborhood of 150 tokens per second, which makes it unusually attractive for code generation, repository-scale ingestion, and other latency-sensitive pipelines where free inference has historically meant slow responses.

In practice, the endpoint is best suited to developers and teams who want to push very large prompts through a text-only pipeline without worrying about meter burn, since the catalog and the routing listing both report zero input, output, and cache-read pricing alongside a 256,000-token context window. The trade-off is moderate raw intelligence relative to paid flagships, which third-party scoring places in the same band as small open models, so the strongest fits are bulk experimentation, retrieval-augmented generation over large document sets, and prototyping where scale of input matters more than peak reasoning depth. The model is currently marked deprecated, which signals that any production deployment should plan a migration path even while the free tier remains useful for cost-free exploration of Xiaomi's MoE design.