Python SDK v0.13.8

Fixed

• Use core 0.3.7, which exports spans with the worker resource so Python traces appear in trace listing, drops sampling parameters for gpt-6 and Claude models that reject them, raises the Claude max_tokens default, and adds the none/low reasoning efforts.
• An Agent on a gpt-6 model (such as openai/gpt-6-luna) no longer sends its default temperature of 0.7. gpt-6 rejects any temperature with a 400, so every agent call to it failed. The gpt-6 family now counts as an OpenAI reasoning model, like gpt-5 and the o-series; an explicit temperature is still sent. Matches TypeScript.
• An Agent on Claude Opus 4.7 and later, Sonnet 5, Opus 5 or Fable no longer sends its default temperature; these models reject it with a 400, so every default Agent call to them failed. @agent(...) agents now get the same default instead of a plain 0.7 that counted as explicit.
• reasoning_effort now reaches the provider. The client passed it to the native binding, which never read it, so OpenAI never received the requested effort; modalities was dropped the same way.
• GenerateResponse.finish_reason is now the provider's stop reason for non-streaming calls instead of always None, so a max_tokens stop is visible.

Added

• Agent(reasoning_effort=...), and ReasoningEffort.NONE / ReasoningEffort.LOW. gpt-6 accepts none/low/medium/high and rejects minimal; gpt-5 accepts minimal.