Cloudflare cuts Clef-flash price, adds Clef-omni for audio and video

Cloudflare updated its Clef family of decision models on Workers AI on 9 October. Clef-flash, the smaller model launched on 1 October, drops from $0.090 to $0.038 per million input tokens, though its hosted context window shrinks from 64,000 to 24,000 tokens, which Cloudflare says affects only 0.24% of requests. The original Clef model keeps its price at $0.240 per million tokens and its 64,000-token context, but gets up to 2x faster after an SGLang integration.

Alongside the price change, Cloudflare launched Clef-omni, a new multimodal decision model built on a 30-billion-parameter mixture-of-experts backbone with 3 billion active parameters. It takes text, images, audio and video in a single request, up to 4 audio clips and 2 videos, and returns a probability for every allowed option of every question, priced at $0.150 per million input tokens. Response times range from about 20 milliseconds for text to about 300 milliseconds for a 21-second video clip with sound.

Why it matters: decision models return a typed probability instead of a block of text, which makes them a cheaper fit for routing, moderation and classification than calling a full LLM. Clef-omni extends that pattern to audio and video, so a moderation or triage pipeline that previously needed a separate multimodal model for media can now use the same decision interface.

The caveat: the Clef-flash price cut comes with a smaller context window, down from 64,000 to 24,000 tokens, so check your prompts still fit before switching.