Batch Results Analyzer
Add your batch request file and result files to see what succeeded, what failed and why. You can then download your results joined to their requests, plus a retry file containing only the requests that need running again.
Runs entirely in your browser. Your file is never uploaded. How this works
What to add
The tool works out what each file is from its contents, so you can add them in any order.
| Provider | Files |
|---|---|
| OpenAI | Your input file, the output file, and the error file if there is one |
| Anthropic | The results file, plus your requests as JSONL with custom_id on each line |
What you get
- Summary: counts of succeeded, failed and missing requests, with total tokens.
- Failure reasons: errors grouped by code, most common first.
joined.jsonl: each request with its result, in your original order.retry.jsonl: only the requests that need running again.
Format details: OpenAI Batch API and Anthropic Message Batches.
Frequently asked questions
- Why do I need to match batch results by custom_id?
- Both OpenAI and Anthropic can return results in a different order from your requests, so line 1 of the output is not necessarily line 1 of the input. The custom_id is the only reliable link.
- Where do failed OpenAI batch requests go?
- OpenAI writes successful requests to the output file and failed ones to a separate error file, found through the batch’s error_file_id. Add both here.
- What does the retry file contain?
- The original request lines for everything that failed, expired, was canceled, or has no result at all, ready to submit as a new batch. Fix errored requests first, since a request that failed validation will fail again.
- Does this estimate cost?
- No. It totals input and output tokens from the results. Prices change and vary by model, so check your provider’s pricing page.