A tool routes returns to shipping or billing. When moving from beta to general availability, the first goal is to confirm that requests still arrive in the correct queue.
What changed on October 22
On October 22, OpenAI made the Decisions API generally available with gpt-6-luna. Changes include shared inputs across requests for caching, images in individual question instructions and direct JSON via input_json. The per-request limit increased from 200 to 400 questions.
What to check in your application
Replay the same fictional requests with the beta configuration and the migrated client. Use archived responses if the old service is unavailable. Compare types and identifiers, misrouted requests and latency. Then separately test batches of 200 and 400 questions and simulate a rejected request or a missing answer.
Switch after validating the task, with a manual queue as fallback. The 400-question cap is a technical limit. A useful batch size depends on your response target and the cost of repeating a batch.