@ai-sdk/openai

Feat (provider/openai): support streaming tool calls that are sent in one chunk

Fix (provider/openai): handle error chunks when streaming

Fix (provider/openai): return unknown finish reasons as unknown

Feat (provider/openai): send user message content as text when possible

Feat (provider): implement toolChoice support

Fix (provider/openai): remove object type validation

Fix (providers): Zod is required dependency

Fix (provider/openai): introduce compatibility mode in which "stream_options" are not sent

Feat (provider/openai): add gpt-4o to the list of supported models

Feat (provider/openai): set stream_options/include_usage to true when streaming

Feat (ai/core): add embed function

Allow optional content to support Fireworks function calling.

ai/core: Add logprobs support (thanks @SamStenner for the contribution)

ai/core: Add support for getting raw response headers.

ai/core: Remove scaling of setting values (breaking change). If you were using the temperature, frequency penalty, or presence penalty settings, you need to update the providers and adjust the setting values.

Add custom header and OpenAI project support.

Simplified model construction by directly calling provider functions. Add create... functions to create provider instances.

Fix build for release.

Support streams with 'chat.completion' objects.

Rename baseUrl to baseURL. Automatically remove trailing slashes.