Clef Flash
4 storiesClef Flash is an open-source decision model from Cloudflare, released on Workers AI alongside Clef. Both are built on Apache 2.0-licensed Qwen and return structured outputs with probabilities so agents can route and decide without generating text. Cloudflare says Clef-flash classifies in about 39 milliseconds, over ten times faster than Jev. In recent community tests, a 9B Q4 version played Google Snake in real time on an RTX 5080, and it placed behind jev 1.13 and GPT-6 Luna on a Pac-Man leaderboard.
Related topics
A Reddit user benchmarked six decision models by having them play Pac-Man in real time and published an open-source leaderboard. jev 1.13 led with an average score of 2,750, followed by GPT-6 Luna and Clef Flash. Users can also play as Pac-Man against model-controlled ghosts.
A 9B Clef Flash model (Q4 quantized) on an RTX 5080 can play Google Snake in real time with no training or game-state hacking, using only simple instructions about what the snake can see. Its decision latency is below the game's 135ms turn limit, approaching human-level play. The approach could extend to other games requiring constant decisions.
Cloudflare launched Clef and Clef-flash, challenging TypeSafe AI's Jev decision model. Clef-flash delivers classifications in about 39 milliseconds, over ten times faster than Jev. Both models are built on Apache 2.0-licensed Qwen and let AI agents make structured decisions without generating text.
Cloudflare released two in-house decision models, Clef and Clef-flash, on Workers AI, saying Clef leads the Jev Decision Index. Clef is open-sourced on Hugging Face under Apache 2.0. Decision models return structured outputs with probabilities so agents can route and decide programmatically; Cloudflare also debuted an RL fine-tuning product.
That is everything