Cloudflare flags overpowered AI model use
Cloudflare has updated its free User Insights tool for AI Gateway to show when teams are sending simple tasks to costlier, more capable models than they need, and which users or agents are responsible.
Cloudflare announced an update to User Insights, the AI Gateway analytics feature it launched a month earlier. The update adds context about the work behind AI traffic, not just model names and request counts. The features are free for AI Gateway users.
The main addition is a "model overkill" view. It flags conversations where the chosen model seems more capable than the task needs, such as simple formatting or summarization requests sent to a high-capability reasoning model. It also shows which users, agents or applications are linked to that pattern. Cloudflare says the view "is not a leaderboard" and does not automatically recommend a replacement model. Instead, it lets teams compare latency, input and output tokens, conversation turns and total cost for the same type of task.
Two other views support it:
- Task analysis sorts conversations by type of work. The first categories are coding, research, writing, summarization and data analysis.
- Turns analysis shows how much back-and-forth a task takes, to capture the full cost of finishing it rather than just the first request.
Cloudflare says these signals also feed a new Potential Savings view, which identifies requests that a faster or cheaper model might handle without losing output quality. They also feed the Auto Router, which launches in public beta with this release and applies the task and model-fit signals automatically, so teams don't need a separate routing rule for each workload.
The post does not say how tasks are classified or how "overkill" is determined. It also gives no details on Auto Router pricing or which models it supports.