
What happened
Cloudflare released Clef and Clef-flash, decision models that return classifications with probabilities instead of text. It reports median latency of about 39 milliseconds for Clef-flash and about 209 milliseconds for Clef, versus just over 524 milliseconds for TypeSafe AI's Jev.
Why it matters
If those latency figures hold in practice, agents could classify a support ticket or a domain and act on the result without a person checking each step, which is what Cloudflare means by saying a human no longer necessarily needs to be in the loop.
What to watch
The speed and accuracy claims are Cloudflare's own, so the test is whether independent users reproduce them. Cloudflare also plans a self-service fine-tuning platform for Clef, after forward deployed engineers handle fine-tuning with customers first.
WHO IT HITSTeams that route support tickets, triage abuse reports, or classify websites could replace slow language-model calls with a faster decision model; Cloudflare is already testing Clef on website classification through its threat intelligence team. Data and platform engineers evaluating agent workflows are the likeliest early users.
Summaries like this, in your inbox every morning.
Cloudflare has mainly been known for its global network for content delivery, DNS, and security services, and its recent AI headlines have centered on giving website owners control over access, such as letting site operators block or allow AI bots based on their purpose in July. Clef marks a shift toward Cloudflare's own AI models, aimed at a niche between large language models, which can reason and call tools but vary in output and can be slow, and traditional classifiers, which are fast but need retraining for every new category.
The approach was popularized by TypeSafe AI, the startup founded by former OpenAI researcher Diogo Almeida, which introduced Jev in mid-September and markets it as a model "without hallucinations" — though that only guarantees it stays within predefined answer options and doesn't prevent it from choosing incorrectly. In late September, OpenAI followed with a Decisions API built on GPT-6 Luna that also accepts context as text or images. Cloudflare keeps its API fully compatible with Jev so customers can switch easily, and it built Clef on Qwen models while leaving the base models unchanged and training extra components on its own synthetic data.
The stakes likely hinge on whether Cloudflare's self-reported benchmarks survive outside scrutiny, and on how quickly the planned self-service fine-tuning platform arrives, since the launch version relies on forward deployed engineers working directly with customers. Buyers weighing a switch from Jev gain from the compatible API, but the real test is whether the latency and accuracy advantages hold on their own tasks rather than on Cloudflare's.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Talkdesk's survey of 252 director-level-and-above leaders found 98% use some form of AI in customer experience…
The Allen Institute for AI announced Olmo-core 3 on Thursday, a training framework it says scaled mixture-of-e…
Ben Thompson wrote that Meta's new Meta Enterprise Platform "won't work" and is "a distraction from the bigges…

Broadcom is reportedly raising $60 billion to fund chips for Anthropic, on top of a loan of up to $42 billion…

Instinct, a year-old San Francisco startup building a personal AI assistant for everyday tasks, raised $1 bill…

At Fortune's AIQ Summit, Honeywell CTO Suresh Venkatarayalu said customers demand 99.9999% accuracy while fron…
