Newer OpenAI models fail with "max_tokens is not supported, use max_completion_tokens instead"

[BoseKarthikeyan]

Summary

LLM analysis fails for newer OpenAI-compatible models (gpt-5+, o1, o3 series) because PAC sends max_tokens in the request body, but these models only accept max_completion_tokens.

PAC Version

v0.39.3

Error

OpenAI API error: Error code: 400 - {'error': {'message': "Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead.", 'type': 'invalid_request_error', 'param': 'max_tokens', 'code': 'unsupported_parameter'}}

Root Cause

In pkg/llm/providers/openai/client.go, the openaiRequest struct hardcodes max_tokens:

type openaiRequest struct {
    Model     string          `json:"model"`
    Messages  []openaiMessage `json:"messages"`
    MaxTokens int             `json:"max_tokens,omitempty"`
}

Newer OpenAI models (gpt-5, gpt-5-mini, gpt-5-nano, o1, o3 series) reject max_tokens and require max_completion_tokens instead.

Proposed Fix
Add MaxCompletionTokens to the request struct and conditionally send the correct field based on the model:
```go
type openaiRequest struct {
    Model               string          `json:"model"`
    Messages            []openaiMessage `json:"messages"`
    MaxTokens           int             `json:"max_tokens,omitempty"`
    MaxCompletionTokens int             `json:"max_completion_tokens,omitempty"`
}

// In Analyze():
apiRequest := &openaiRequest{
    Model:    c.config.Model,
    Messages: []openaiMessage{{Role: "user", Content: fullPrompt}},
}
if isNewGenerationModel(c.config.Model) {
    apiRequest.MaxCompletionTokens = request.MaxTokens
} else {
    apiRequest.MaxTokens = request.MaxTokens
}

func isNewGenerationModel(model string) bool {
    newGenPrefixes := []string{"gpt-5", "o1", "o3"}
    for _, prefix := range newGenPrefixes {
        if strings.HasPrefix(model, prefix) {
            return true
        }
    }
    return false
}

## Additional Context

This also affects custom OpenAI-compatible API gateways that proxy newer models. The omitempty tag on max_tokens does not help because PAC applies a default value of 1000 when max_tokens is not set in the Repository CR.