这是中转查对 cap-ignored 的协议面与可用性快照。 它只回答“能不能用、协议面是否像官方”,不认证型号身份。
主分只看稳定性、对话、能力、目录;协议契约、错误码和包装在附录,不计入主分。
| 目标 | 型号 | 稳定性 | 对话 | 能力 | 目录 | 总分 |
|---|---|---|---|---|---|---|
| cap-ignored | gpt-6.1-sol | pass 100% (3/3) | pass 100% (4/4) | pass 100% (4/4) | pass 100% (1/1) | A |
主分只看能不能用:稳定性、对话、能力、目录;协议细节在下面附录,不计入主分。
| 目标 | 包装 | 协议契约 | 错误码 |
|---|---|---|---|
| cap-ignored | n/a | fail 75% (3/4) | pass 100% (3/3) |
| 检查 | 结果 | 说明 |
|---|---|---|
responses.surface | pass | Responses surface returned a completed response |
responses.attribution | pass | usage.attribution present |
responses.reasoning_items | info | no reasoning item in this short call |
responses.final_answer | pass | message phase=final_answer |
wrapper.instructions | info | no injected instructions |
reasoning.scaling | pass | reasoning tokens increase with effort |
previous.roundtrip | pass | previous_response_id roundtrip works |
invalid.parameter | pass | invalid reasoning.effort rejected as invalid_request_error |
error.invalid_model | pass | invalid model rejected with an official-looking error |
error.invalid_cap | pass | max_output_tokens=0 rejected as invalid_request_error |
error.unauthorized | pass | missing credential rejected with an auth error |
image.vision | pass | image input answered correctly |
chat.surface | pass | Chat Completions surface works |
catalog.liveness | pass | all sampled catalog models answered |
usage.stability | pass | input_tokens stable across repeats |
store.echo | pass | store flag echoed as expected |
cache.retention | pass | prompt_cache_retention accepted |
text.verbosity | pass | text.verbosity echoed as low |
tools.function_call | pass | Responses function call produced with expected args |
structured.output | pass | JSON schema output matched |
identity.undetermined | info | protocol surface only; model identity is not authenticated |
output.cap | fail | max_output_tokens was ignored (returned more tokens than the cap) |
源目录:examples/scenarios/cap-ignored
文件:PLAN.json, RESULTS.json, SUMMARY.json