Is the Batch API on Bedrock, Vertex or Foundry?

The mistake

Treating "Batch isn't available on Bedrock, Vertex AI or Foundry" as an established fact. This page asserted it, and this site's reference tables listed all three platforms. At the October re-check no OpenAI page said it any more — and two of the three platforms aren't mentioned on OpenAI's pages at all.

Why this happened

The claim came from this site's first research pass. Re-checking it, the Batch guide no longer names any platform, the pricing page now refers only to Amazon Bedrock and Microsoft Azure as platforms that bill through their own services, and OpenAI's Bedrock guide defers service tiers to AWS's own pricing page and model cards. A negative claim with no current source is exactly what this site's rules forbid, so it has been withdrawn rather than left standing on the strength of an old reading.

Why it matters

The underlying risk the old page warned about is real; only its certainty was wrong. If a cost model depends on Batch's 50% discount and the deployment runs through a cloud platform, you don't know the discount exists there until that platform's own documentation says so. Discovering otherwise after the integration is built means paying the Standard rate on every request you planned to batch — and since moving a workload between platforms is an architecture decision, not a config change, it's expensive to fix late.

The fix

Before designing around Batch on a cloud platform, find Batch — or an equivalent asynchronous discount — in that platform's own pricing documentation, and price the fallback case at Standard rates. If the discount is essential and the platform doesn't offer it, calling OpenAI directly is the reliable route; weigh that against what the platform provides, such as consolidated billing or residency.

What to ask a platform before relying on it

A few questions settle it: whether the platform offers an asynchronous batch mode for the OpenAI model you plan to use, what discount it applies and to which token types, and whether its jobs run against limits separate from your live traffic — on OpenAI direct, Batch draws on its own pool, which is much of what makes it useful for large jobs. If the platform's documentation doesn't answer one of these, treat it as unconfirmed rather than assumed.