OpenAI Batch API Validator

updated 28 July 2026

Validate a Batch API input file before you upload it. Checks every line for a unique custom_id, the required method, a supported url, and a well-formed body — plus the file-level 50,000-request and 200 MB limits. Runs in your browser.

What gets checked

• custom_id — present, a non-empty string, at most 64 characters, and unique across the file. Duplicates are the most common reason a batch is rejected, and the report names the earlier line so you can find the clash.
• method — must be exactly "POST".
• url — one of /v1/chat/completions, /v1/completions, /v1/embeddings, /v1/responses. Anything else is a warning by default, an error in strict mode.
• One endpoint per file — if several url values appear, the report says so. A batch cannot mix endpoints.
• body — an object containing model. For chat completions, messages must be a non-empty array whose entries have a recognised role and either content or tool_calls. For embeddings, input is required.
• stream — must not be true; batch requests cannot stream.
• File limits — 50,000 requests and 200 MB, with a warning once you pass 150 MB.
• Unexpected top-level keys — reported as warnings, since they are ignored by the API and usually indicate a generation bug.

Batch format vs fine-tuning format

These two JSONL formats look similar and are not interchangeable — a mistake that costs people a rejected upload surprisingly often.

• Fine-tuning files hold training examples: each line is {"messages": […]} with no wrapper. Validate those with the OpenAI fine-tune validator.
• Batch files hold API requests: each line wraps a full request body in custom_id / method / url / body. That is this page.

The practical tell: if your lines start with {"custom_id" you want this validator; if they start with {"messages" you want the fine-tune one.

After validation

Use Token counter to estimate the cost before you submit — batch pricing is discounted but 50,000 requests still add up. Splitter breaks an over-limit file into compliant chunks, and Filter pulls out just the requests you want to retry.

Privacy

Nothing is uploaded. The whole thing runs in this tab, in your own browser. That matters here more than on most tool sites — training data is usually the most sensitive file a team owns.

Frequently asked questions

Related tools

• OpenAI fine-tune validatorFor training files, not batch requests.
• Token counterEstimate the cost of a batch before submitting.
• SplitterBreak an over-limit file into compliant chunks.
• Encoding fixerStrip invisible characters that break id uniqueness.