fix

apps/api/internal/processing/openai_chat_params.go

@@ -0,0 +1,97 @@
+package processing
+
+import (
+	"strings"
+)
+
+// openAIChatUsesMaxCompletionTokens reports models that reject legacy max_tokens
+// on Chat Completions (o-series + GPT-5 family including GPT-5.6).
+// See https://developers.openai.com/api/reference/resources/chat
+func openAIChatUsesMaxCompletionTokens(model string) bool {
+	return openAIReasoningChatModel(model)
+}
+
+// openAIChatOmitsCustomTemperature reports models that reject non-default
+// temperature (only the API default of 1 is accepted, if temperature is sent).
+func openAIChatOmitsCustomTemperature(model string) bool {
+	return openAIReasoningChatModel(model)
+}
+
+// openAIChatSupportsReasoningEffort reports Chat Completions models that accept
+// top-level reasoning_effort (Responses API uses reasoning.effort instead).
+func openAIChatSupportsReasoningEffort(model string) bool {
+	return openAIReasoningChatModel(model)
+}
+
+// openAIReasoningChatModel matches OpenAI reasoning / GPT-5(+).5/5.6 chat models
+// and common snapshot aliases (e.g. gpt-5.6-luna-2026-…).
+func openAIReasoningChatModel(model string) bool {
+	m := strings.ToLower(strings.TrimSpace(model))
+	if m == "" {
+		return false
+	}
+	// Strip provider prefixes used by OpenRouter-style gateways.
+	if i := strings.LastIndex(m, "/"); i >= 0 {
+		m = m[i+1:]
+	}
+	switch {
+	case strings.HasPrefix(m, "o1"),
+		strings.HasPrefix(m, "o3"),
+		strings.HasPrefix(m, "o4"),
+		strings.HasPrefix(m, "gpt-5"),
+		strings.HasPrefix(m, "chatgpt-o"):
+		return true
+	default:
+		return false
+	}
+}
+
+// defaultReasoningEffort picks Chat Completions reasoning_effort when unset.
+// Long formula-HTML enhance uses "low" so max_completion_tokens prefers visible
+// HTML over deep hidden reasoning. Tiny probes use "none" for speed/cost.
+func defaultReasoningEffort(model string, maxTokens int, explicit string) string {
+	explicit = strings.ToLower(strings.TrimSpace(explicit))
+	if explicit != "" {
+		return explicit
+	}
+	if !openAIChatSupportsReasoningEffort(model) {
+		return ""
+	}
+	if maxTokens <= 0 || maxTokens < 512 {
+		return "none"
+	}
+	return "low"
+}
+
+// chatCompletionBody is the Chat Completions JSON body. Temperature is a pointer
+// so reasoning models can omit it entirely (omitempty).
+type chatCompletionBody struct {
+	Model               string        `json:"model"`
+	Messages            []chatMessage `json:"messages"`
+	Temperature         *float64      `json:"temperature,omitempty"`
+	MaxTokens           int           `json:"max_tokens,omitempty"`
+	MaxCompletionTokens int           `json:"max_completion_tokens,omitempty"`
+	ReasoningEffort     string        `json:"reasoning_effort,omitempty"`
+}
+
+// buildChatCompletionBody maps CompleteOptions onto the correct Chat Completions
+// fields for classic vs GPT-5 / o-series models.
+func buildChatCompletionBody(model string, messages []chatMessage, temperature float64, maxTokens int, reasoningEffort string) chatCompletionBody {
+	body := chatCompletionBody{
+		Model:    model,
+		Messages: messages,
+	}
+	if openAIChatUsesMaxCompletionTokens(model) {
+		if maxTokens > 0 {
+			body.MaxCompletionTokens = maxTokens
+		}
+	} else if maxTokens > 0 {
+		body.MaxTokens = maxTokens
+	}
+	if !openAIChatOmitsCustomTemperature(model) {
+		t := temperature
+		body.Temperature = &t
+	}
+	body.ReasoningEffort = defaultReasoningEffort(model, maxTokens, reasoningEffort)
+	return body
+}