Umans Flash

Umans AI Coding Plan is documented as an OpenAI-compatible provider that routes requests through an API endpoint, with its catalog enumerated on third-party model directories. The visible roster on models.dev includes derivatives built on base models from DeepSeek, Zhipu (GLM), and Moonshot (Kimi), each carrying distinct context windows and per-token pricing rather than a uniform free tier. Mastra's router documentation confirms seven Umans AI models are reachable via the OpenAI-compatible /chat/completions path, though the supplied excerpt only illustrates one identifier in code and does not enumerate every slot.

Within the evidence supplied, no model entry named Umans Flash or keyed as umans-flash appears in either the models.dev table or the Mastra router listing, so claims about a specific Qwen-family origin, open-weights status, multimodal input handling, capability profile, zero-cost pricing, or 262,144-token limits cannot be substantiated from these sources. The catalog's qwen family attribution also diverges from the visible lineup of DeepSeek, GLM, and Kimi derivatives. A model with this identifier may exist beyond the truncated portions of the supplied excerpts, but the present evidence is insufficient to characterize its architecture, training lineage, benchmark performance, or practical strengths.