It also says Clef inference is up to 2.0x faster and Clef-flash pricing has been lowered.
The thesis is that the Clef announcement pairs a broader input surface with changes to inference speed and price, but the evidence supplied establishes claims rather than comparative performance. The mechanism is a unified model pipeline for multiple input types announced alongside separate speed and price adjustments. Why now: Cloudflare is presenting the multimodal model and the speed and pricing changes together as an expansion of the Clef family. This combination could reduce the need to route different media types through separate models, while lower Clef-flash pricing could affect deployment choices; these are possible implications, not reported customer results.
The evidence boundary is the company announcement; the supplied material includes no detailed pricing table, workload-level speed methodology, or independent evaluation. The first implication is that teams evaluating Clef can consider multimodal input, inference speed, and Clef-flash cost together, while verifying fit against their own workloads. The first watch condition is whether Cloudflare publishes comparable benchmarks, exact pricing, and practical limits for the supported modalities. No prior New Runtime record was supplied for this announcement, so this starts a distinct analytical record.