GPT-5 does not work well compared to 4.1

[sibblegp]

Describe the bug

A clear and concise description of what the bug is.

Debug information

• Agents SDK version: 0.2.4
• Python version 3.13

Repro steps

I've been using GPT-5 as a drop in for GPT-4.1 for the last hour. It's a disaster. While 4.1 understood my requests, routed to the right agent, and called tools, GPT-5 keeps asking for more specifics and details before it'll call a tool. It's a much worse experience.

Like, for example, I sent in "What is the weather tomorrow night?"

My system prompt has a location in it for each request as well as the current date time and time zone. There is a tool that takes latitude and longitude and a date and gets the hourly forecast.

ChatGPT 4.1 does this 100% correctly. Gets the hourly forecast and gives a great answer.

GPT-5 asks if I mean tomorrow night (with the date) specifically and doesn't call the tool without more input.

It even sometimes say it'll do something but doesn't actually call the tool.

This is an unmitigated disaster and GPT-5 doesn't work at all for agents.

It's also very slow at handing off while actual responses do come back quickly. 4.1 overall is faster.

Expected behavior

GPT-5 should outperform 4.1 in terms of tool calling but it understand commands and tools less, fails to take the correct actions, and is slower than 4.1

[sibblegp]

Two examples. These are queries that 4.1 got right 100% of the time.

It should know that I mean tomorrow night, Tuesday. Not sure why it's asking for clarification.

In this one it transferred to the agent, but didn't actually make the tool call. This is literally just broken. Again, 4.1 nails this.

[sibblegp]

Compare to 4.1's handling of the same question.

Clear, concise and fast.

[sibblegp]

And for asking for a flight status:

[moseshu]

I got same questions.
gpt5 can't support function calls

openai.NotFoundError: Error code: 404 - {'error': {'message': 'tools is not supported in this model.

but this (https://platform.openai.com/docs/guides/function-calling#models-supporting-function-calling)
shows that function call is supported

[ariveram2111]

I am seeing other errors with GPT-5 too as

{'error': {'message': "Missing required parameter: 'input[1].id'.", 'type': 'invalid_request_error', 'param': 'input[1].id', 'code': 'missing_required_parameter'}}

And the model is not calling tools

[bakikucukcakiroglu]

I also experience extreme slowness. gpt-5 takes ~1 minute for the same query that gpt-4.1 answers in 5 seconds.

I think this is related to Responses API. Please see OpenAI Developer Forum: GPT-5 + Responses API is extremely slow

[sibblegp]

GPT-5 in general seems to break tool calls, which is odd because that was a solved problem in 4.1

@seratch @rm-openai This is a pretty urgent issue if you want GPT-5 to be adopted.

[seratch]

We appreciate this feedback, and we're aware of the issue. We will look into viable solutions, but for now, please continue using gpt-4.1 as the default model for most use cases.

[sibblegp]

@seratch Appreciate the acknowledgement. Let us know when it gets resolved.