Cloudflare has entered the AI market by developing its own specialized models, positioning itself as a neolab. The company aims to optimize operational tasks by deploying small, efficient models designed for classification and data extraction rather than open-ended text generation.
The company launched the Clef and Clef-flash models on October 1, 2026. Clef features 27 billion parameters, while Clef-flash contains 9 billion. Both models support vision inputs and handle context windows of up to 64,000 tokens, with Clef-flash achieving a median latency of approximately 39 ms in internal tests.
On October 19, 2026, Cloudflare introduced Clairic, an extraction model priced at $0.10 per million input tokens. This model is designed to pull specific data points from unstructured inputs, complementing the existing Clef lineup to handle triage tasks before engaging larger, more expensive language models.
Cloudflare leverages its existing edge computing infrastructure and distributed GPUs to run these models closer to end users. By offloading routine tasks to smaller models, the company claims it can significantly reduce token usage, citing a test case where a diagnosis task was 3.5 times faster and 77% cheaper than using a large model alone.




