Skip to content

v2.7.0: GPT-6 Luna migration

Choose a tag to compare

View all tags
@github-actions github-actions released this 24 Sep 00:43
· 34 commits to main since this release
v2.7.0
176670a
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

GPT-6 Luna for routine data processing

The general OpenAI Responses default is now gpt-6-luna (previously gpt-4.1-mini). Current documentation and examples use explicit reasoning={"effort": "none"} for routine classification, extraction, translation, and table processing. GPT-6 Sol remains an explicit option for more demanding tasks.

Upgrade notes

  • Omitting reasoning preserves SDK semantics and uses Luna's medium default. With reasoning other than none, remove sampling/logprobs options that the model does not support.
  • Existing applications can pin openaivec.set_responses_model("gpt-4.1-mini") while evaluating the new default.
  • Recreate wrappers and registered Spark/DuckDB UDFs after configuration changes. Use separate caches for each model, prompt/schema, and request configuration.
  • Fabric built-in models remain controlled by Fabric's release cycle. Their gpt-5.1, text-embedding-ada-002, and API-version defaults are unchanged. Azure deployment identifiers and the general embeddings default are also unchanged.
  • Historical benchmark measurements are retained as historical results; they are not GPT-6 performance claims.

Fixes

  • Schema inference now refreshes after configuring a different client or Responses model, while preserving custom providers and caller-owned client lifetimes.
  • Explicit RetryPolicy honors numeric and HTTP-date Retry-After. If the requested delay exceeds max_delay, the original API error is propagated; otherwise, a delay reaching the shared deadline raises TimeoutError. SDK-owned retries are unchanged.
  • Corrected request-option placement in examples and enabled the intended pytest configuration.

See the migration guide and PR #195.

Validation

GPT-6 live migration checks passed on OpenAI SDK 2.0.0, 3.8.0, and 3.13.0. Offline tests, Ruff, source type checking, wheel/sdist builds, notebook validation, and the documentation build passed. The full GitHub Actions suite passed 1010 tests, including live API tests; CodeQL and CLA checks passed.

pip install --upgrade openaivec==2.7.0
Assets 6
Loading