Text models
GPT-6 Astra integration checks
Check SDK behavior, tool calls, usage, and regressions against a smaller model.
Validate the endpoint contract
Start with a short response, then check streaming and a round trip through a tool call. Keep the model ID, response fields, and request ID from each run.
Compare your workload
Include a code change, a long log, and a structured extraction. Use the same pass criteria and compare quality, runtime, and the final charge.
Test prompt
Read the files and fix the failing test. Show the reproduction command, the smallest patch, and verification output.