AI Answer stops working on 23 October: the gpt-3.5-turbo default shuts down, and its replacement rejects max_tokens

[Lakshya-0910]

When no model is set, the AI Answer pipeline stage and the OpenAI plugin default to gpt-3.5-turbo with 300 max tokens (plugins/pipelines/ai_answer.go:109-122, plugins/openai/util.go:30-32). OpenAI shuts down gpt-3.5-turbo on 23 October 2026 and names gpt-5.6-terra. BuildChatGPTBody always sends the limit as max_tokens (plugins/openai/util.go:1267), and terra replies:

Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead.

The same applies to anyone who sets a gpt-5* model in their config today, since max_tokens is always sent.

Fix — either one (both tested live with your default prompt):

• One line each: default to gpt-4.1-mini (not retiring; accepts max_tokens) in ai_answer.go:109 and util.go:30.
• Support the replacement: send the limit as max_completion_tokens for gpt-5* models. With 300, terra answered normally.

 	// If maxTokens is passed, inject it in the request body
 	if maxTokens != nil {
-		requestBodyAsMap["max_tokens"] = *maxTokens
+		if strings.HasPrefix(model, "gpt-5") {
+			requestBodyAsMap["max_completion_tokens"] = *maxTokens
+		} else {
+			requestBodyAsMap["max_tokens"] = *maxTokens
+		}
 	}

[nisatsama]

I'd like to work on this issue. I'll investigate the existing OpenAI request construction, determine the most compatible fix for gpt-5* models while preserving existing behavior, and add the necessary regression tests.

[Lakshya-0910]

Thanks for offering, @nisatsama! Since I had the fix ready and tested live from writing this issue, I've opened it as #403 so it's in before the 23rd. Reviews welcome.