AIChart not working for gpt-5, and Claude claude-opus-4-1-20250805

[wxinix (Wu)]

AIChart demo don't work with latest OpenAI gpt-5 and Claude claude-opus-4-1-20250805. It works nicely with gpt-4o and claude-sonnet-4-20250514

Also, the request payload generated for gpt-5 is:

{
"model": "gpt-5",
"input": [
{"role":"user","content":"Can you make a bar chart of the population of the 10 biggest countries in Europe"}
],
"tools": [
{ "type": "function", "name": "clearseries", "description": "Clear the series on the chart" },
{ "type": "function", "name": "addbarseries", "description": "Add a bar series to the chart" },
{ "type": "function", "name": "addpieseries", "description": "add a pie series to the chart" },
{ "type": "function", "name": "addpoint", "description": "add a point to a series in the chart", "parameters": { ... } }
],
"max_output_tokens": null
}

ChatGPT suggests to add a new field parallel_tool_calls to the payload - is it possible?

{
"model": "gpt-5",
"input": [
{"role":"user","content":"Can you make a bar chart of the population of the 10 biggest countries in Europe"}
],
"tools": [
{ "type": "function", "name": "clearseries", "description": "Clear the series on the chart" },
{ "type": "function", "name": "addbarseries", "description": "Add a bar series to the chart" },
{ "type": "function", "name": "addpieseries", "description": "add a pie series to the chart" },
{ "type": "function", "name": "addpoint", "description": "add a point to a series in the chart", "parameters": { ... } }
],
"parallel_tool_calls": true,
"max_output_tokens": null
}

[brunofierens]

We will investigate.
Meanwhile, you can try to use the Settings.CustomOptions to set it to

"parallel_tool_calls":true,

[wxinix (Wu)]

thank you for the prompt reply. It turns out, adding the following system role would make AIChart demo work with gpt-5, and claude-opus-4-1-20250805:

"When returning function calls, never repeat the same function_call ID more than once in subsequent inputs. After a function call has an output, include only the function_call_output, not the raw function_call again."

[brunofierens]

Still weird to have to resort to this kind of solution. We'll investigate deeper.