I came across an example in the “GPT-4.1 Prompting Guide”, which recommends using “Prompting-Induced Planning & Chain-of-Thought” techniques. The provided SWE-bench Verified example uses the responses API, and demonstrates how the model can reason step-by-step before invoking a tool. Here’s a snipp…

Here is a thought experiment for you. Which of these is more likely for the AI to write? Case 1 (tool call) get_weather(“location”: “San Diego”} “I invoked the weather tool, I’m now waiting for it to return results so I can tell you the weather” Case 2 “I’ll use my weather tool to retrieve the co…

I believe the output list is in chronological order.

I’m wondering what OpenAI does differently in the Responses API that allows GPT-4.1 to write preamble text and then call a tool. When I use the same GPT-4.1 model and the same prompt via the Chat Completions API, the model only generates the preamble text—but doesn’t actually include the tool call i…

How can I implement CoT reasoning before tool calling using the Chat Completions API?

API

Dimitrius June 23, 2025, 2:57pm 2

I have a similar issue when Responses return multiple messages + function_call. Read my thread

My guess is that the output list is prioritized, so the most relevant answer is the first one.

Topic		Replies	Views
Force communication through tool API tool , function-calling , agents , assistants-api	4	568	November 22, 2024
Can prompt design enhance model's planning/reasoning when using function calling? API functions	13	4523	March 30, 2024
How can I ensure every LLM reply includes exactly one message and one tool call? Prompting tools , gpt-41	9	1578	June 20, 2025
How to include content and tool calls in one message? API gpt-4	1	2738	October 16, 2024
Getting a function call + textual response in the same call API gpt-4 , function-calling	9	2933	April 8, 2025

How can I implement CoT reasoning before tool calling using the Chat Completions API?

Related topics