Cloudflare’s User Insights now flags when you’re using too much AI model

Cloudflare has added a “model overkill” view to User Insights, the AI Gateway analytics tool it launched last month. The update, announced on the company’s blog on September 30, 2026, is free for AI Gateway users and is meant to show when a selected model appears more capable than the task requires.

According to Cloudflare, the feature does not name a better model or rank users. Instead, it points to conversations where a high-capability reasoning model is handling something simple, such as formatting or summarization, and shows which users, agents, or applications are behind that pattern.

What the overkill view actually shows

Cloudflare says model names and request counts only tell part of the story: the same token count can represent a code review, a research task, or an agent making several calls to finish a job. The new view ties task, model, cost, and conversation patterns together so teams can investigate before changing a model, workflow, or routing rule.

The company lists the questions the view is designed to prompt: whether the model fits the task, whether the extra capability improves the result, whether a faster or cheaper model would produce an equivalent outcome, and whether the issue is limited to one workflow, user, or agent. Teams can then compare cost, latency, token usage, and conversation turns.

Potential Savings and Auto Router

The overkill data also feeds two related features. The Potential Savings view, per Cloudflare, highlights requests that may be handled by a faster or less expensive model without hurting output quality. The Auto Router applies those task and model-fit signals automatically, so teams do not need a separate routing rule for every workload.

Cloudflare says the Auto Router is launching in public beta alongside this release. The company describes the overkill view as a starting point rather than a leaderboard, and notes that a model may be in use simply because it is the default, because users are unsure which model to choose, or because an agent is configured to use the same model for every step.

The new capabilities are available for free to AI Gateway users, according to Cloudflare.

Related reading

Sources

• Identify AI model overuse with User Insights — Cloudflare Blog