Which models support parallel Tool Calling

[nikita1503]

The documentatoin is on parallel tool calling is insufficient. I need help understanding which models support parallel tool calling and for models that support it, is it enabled by default?

[MukundaKatta]

Short version: gpt-4 (from 0613 on), gpt-4-turbo, gpt-4o/gpt-4o-mini, gpt-3.5-turbo (1106 on), and the newer reasoning models o3-mini/o4-mini support it. Controlled by parallel_tool_calls: bool on the request, defaults to true on supporting models. On non-supporting ones the param is silently ignored. Docs: https://platform.openai.com/docs/guides/function-calling#parallel-function-calling

[nikita1503]

Thanks @MukundaKatta for answering.

• Is parallel tool calling enabled for the same models in Response API and Completions API?

• Is there any documentation on how to consume parallel tool calls in Response API?
  On the Responses API (/v1/responses), parallel is still supported but the response shape is different, so downstream parsing code may need adjustment

[MukundaKatta]

Yes, same model list applies on the Responses API. The consumption shape differs though: on /v1/responses you get tool calls as function_call items in the response.output array (non-streaming), or as response.output_item.added / response.function_call_arguments.delta events when streaming. Each parallel call gets its own item with its own call_id, which is what you pair with the corresponding function_call_output on the next turn. Ref: https://platform.openai.com/docs/guides/function-calling?api-mode=responses

[nikita1503]

This resolves my doubt. Thanks :)