Results / request.max-tokens-field
Output limit uses the supported Chat Completions field
A short answer is requested with an output limit that must use max_completion_tokens.
Bug class: The library sends something the caller didn't ask for
The library adds, drops or renames request fields the caller didn't set, or sends a combination the provider rejects.
Calls fail on current models, or behave differently from what the caller configured.
Requirements
| Level | Id | Check | Requirement | Providers |
|---|---|---|---|---|
| MUST | limit-field | request_field_equals | The request sends the output limit as max_completion_tokens. | All |
| MUST | no-legacy-limit | request_field_absent | The request omits the unsupported max_tokens field. | All |
| SHOULD | no-temperature | request_field_absent | The request leaves temperature unset. | All |
Fixtures
| Provider | Tier | any-llm | litellm | pydantic-ai | tanstack-ai | vercel-ai |
|---|---|---|---|---|---|---|
| openai-chat | recorded | ✓ PASS | ✓ PASS | ✓ PASS | – UNSUPPORTED | ✓ PASS |