Model overview
What this model is for
Clef-Flash is a 9B multimodal decision model intended for structured classification, routing, and other latency-sensitive selection tasks. It accepts text, JSON, images, and video state, and produces probability scores for schema-defined choices, yes/no questions, and ordered scores. The model is post-trained from Qwen3.5-9B and exposes a Jev/SystemOne-compatible interface, making it potentially useful for typed decisions, coding-related classification workflows, and vision-based state evaluation. Its listed modalities are text, image, and video, but the supplied source does not disclose a context-window size, inference speed, benchmark results, or pricing. The model card content and update date are not exposed in the supplied Hugging Face HTML, so implementation details and current availability should be verified on the source page before deployment. Open weights are provided under the Apache-2.0 license. Clef-Flash is positioned as a smaller model for latency-sensitive use, but no measured latency or throughput is reported. It should therefore be evaluated against your own routing, decision-quality, and multimodal input requirements rather than selected on an assumed performance advantage.