[Bug] gpt-5.6-sol returns HTTP 500 after shell + custom function continuation

API
, ,
1

I am seeing a reproducible HTTP 500 with gpt-5.6-sol on the Responses API when a response uses the hosted shell tool and then calls a developer-defined function.

The initial request completes successfully and returns these output item types:

[
  "shell_call",
  "shell_call_output",
  "function_call"
]

I then continue using the returned previous_response_id and call_id, supplying a normal string function_call_output:

{
  "type": "function_call_output",
  "call_id": "call_...",
  "output": "{\"ok\":true}"
}

The continuation also includes another ordinary function tool. Instead of accepting the function result and continuing, the API returns:

HTTP 500
type: server_error
code: server_error

Single-file reproduction using the official JavaScript SDK:

Environment

  • Model: gpt-5.6-sol
  • API: Responses API
  • OpenAI JavaScript SDK: openai@6.48.0
  • Node.js: v22.23.1
  • store: true
  • Observed: July 21, 2026

Latest request IDs

  • Successful setup request: req_7fd7e98235fe490baea746f857460797
  • Failing continuation: req_f0e8b1a570d94f26a1768692475d5ac6

Is this a known issue with continuing a response that contains both hosted-shell output and a developer-defined function call?

2

This was resolved by OpenAI support. Thanks!