Responses API Bugs - duplicate output messages and tool calls

[oliver.drobnik]

I encountered the following unexpected results. I am asking for a structured response like the web search planner. I added local tool call to get the current date, because I found that it would plan for latest as of 2023.

I would expect for it to have one tool call to getCurrentDateTime, and then one response with the valid JSON. Usually I also got that, but when running it multiple times in a row, I also got this IMHO incorrect behavior

1. Multiple text output items with JSON: In trace_99B5F1E98F0F4AB1AD26C18D81EA5F09

2. superfluos multiple extra tool calls in addition to desired output: in trace_05D7433E066F44A69E8F443CB67FE44B

I am rolling my own agent framework in Swift, that’s how I noticed that because the Agents SDK hides this misbehavior because it stops the agent run loop when a suitable response is received, and only taks the first response. In other words it never sees these two issues.

Cheers
Oliver

[Maciej_Kiernarzycki]

Same situation here, but i have encountered that on 4.1-mini. This happens in Chat Completions API too.

[URGENT] GPT-4.1-mini consumes unknown amount of input tokens

Today I noticed a very worrying behaviour of 4.1-mini. I sent 119 input tokens, but receive almost 40x more input tokens in response usage. Two short messages and almost 5k input tokens. Response contains multiple duplicated structured outputs with the same message identifier, which also counts as output usage. This is serious issue that can cost someone in production quite a lot. To reproduce that behaviour you need to configure tools, structured output and drop a unresolvable message. Her…

[Maciej_Kiernarzycki]

Also notice the input token spike on the first ss I bet that request content is not almost 500 input tokens.