Workers AI - Reject busy synchronous inference requests
Cloudflare's Developer Platform changelog includes an entry dated 2026-09-17 titled "Reject busy synchronous inference requests," listed under Workers AI. The URL path is /changelog/post/2026-09-17-reject-if-busy/, and the page is labeled as a changelog post. Beyond the title, date, product area, and URL, the supplied source text offers no substantive description: the body consists of embedded CSS rules, layout classes, and styling for changelog pages, code blocks, file trees, and aside cards rather than prose about the change itself. As a result, the source does not state which request types are affected, what error or status code callers receive, whether the behavior is opt-in or default, or how developers should configure or handle it. The only source-grounded claim is the headline behavior: synchronous inference requests to Workers AI can be rejected when the system is busy. Any further detail about thresholds, retry semantics, or rollout would go beyond what the excerpt supports, so it is omitted here.
Developers calling Workers AI synchronously may see requests rejected when the system is busy, per the changelog title.