Merge pull request #2173 from dgageot/board/fix-https-github-com-docker-docker-agent-3744de66

fix: skip reasoning.summary for o1-pro and gpt-5-chat models

Author: dgageot

pkg/model/provider/openai/client.go

@@ -375,8 +375,13 @@ func (c *Client) CreateResponseStream(
 	// Skip reasoning configuration entirely if thinking is explicitly disabled (via /think command)
 	thinkingEnabled := c.ModelOptions.Thinking() == nil || *c.ModelOptions.Thinking()
 	if isOpenAIReasoningModel(c.ModelConfig.Model) && thinkingEnabled {
-		params.Reasoning = shared.ReasoningParam{
-			Summary: shared.ReasoningSummaryDetailed,
+		// Only set reasoning.summary for models that support it.
+		// Some reasoning models (e.g. o1-pro) reject this parameter.
+		if supportsReasoningSummary(c.ModelConfig.Model) {
+			params.Reasoning = shared.ReasoningParam{
+				Summary: shared.ReasoningSummaryDetailed,
+			}
+			slog.Debug("OpenAI responses request configured with reasoning summary", "model", c.ModelConfig.Model, "summary", "detailed")
 		}
 		// Apply thinking budget as reasoning effort if configured
 		if c.ModelConfig.ThinkingBudget != nil {
@@ -388,7 +393,6 @@ func (c *Client) CreateResponseStream(
 			params.Reasoning.Effort = shared.ReasoningEffort(effort)
 			slog.Debug("OpenAI responses request using thinking_budget", "reasoning_effort", effort)
 		}
-		slog.Debug("OpenAI responses request configured with reasoning summary", "model", c.ModelConfig.Model, "summary", "detailed")
 	}
 
 	// Apply structured output configuration
@@ -903,12 +907,33 @@ func isResponsesModel(model string) bool {
 
 func isOpenAIReasoningModel(model string) bool {
 	m := strings.ToLower(model)
+
+	// gpt-5-chat variants are non-reasoning chat models.
+	if strings.HasPrefix(m, "gpt-5-chat") {
+		return false
+	}
+
 	return strings.HasPrefix(m, "o1") ||
 		strings.HasPrefix(m, "o3") ||
 		strings.HasPrefix(m, "o4") ||
 		strings.HasPrefix(m, "gpt-5")
 }
 
+// supportsReasoningSummary returns true for reasoning models that support the
+// reasoning.summary parameter. Some reasoning models (e.g. o1-pro) do not
+// support it and will reject the request if it is set.
+func supportsReasoningSummary(model string) bool {
+	if !isOpenAIReasoningModel(model) {
+		return false
+	}
+	m := strings.ToLower(model)
+	// o1-pro does not support reasoning.summary.
+	if strings.HasPrefix(m, "o1-pro") {
+		return false
+	}
+	return true
+}
+
 // getOpenAIReasoningEffort resolves the reasoning effort value from the
 // model configuration's ThinkingBudget. Returns the effort (minimal|low|medium|high) or an error
 func getOpenAIReasoningEffort(cfg *latest.ModelConfig) (effort string, err error) {

pkg/model/provider/openai/thinking_budget_test.go

@@ -40,6 +40,11 @@ func TestIsOpenAIReasoningModel(t *testing.T) {
 		{"gpt-5-turbo", "gpt-5-turbo", true},
 		{"GPT-5 uppercase", "GPT-5", true},
 
+		// GPT-5-chat variants are non-reasoning chat models
+		{"gpt-5-chat-latest", "gpt-5-chat-latest", false},
+		{"gpt-5-chat", "gpt-5-chat", false},
+		{"GPT-5-CHAT uppercase", "GPT-5-CHAT-LATEST", false},
+
 		// GPT-4 series models - should NOT support reasoning
 		{"gpt-4", "gpt-4", false},
 		{"gpt-4o", "gpt-4o", false},
@@ -51,6 +56,11 @@ func TestIsOpenAIReasoningModel(t *testing.T) {
 		{"gpt-3.5-turbo", "gpt-3.5-turbo", false},
 		{"gpt-3.5-turbo-16k", "gpt-3.5-turbo-16k", false},
 
+		// O1-pro series models - should support reasoning
+		{"o1-pro", "o1-pro", true},
+		{"o1-pro-2025-03-19", "o1-pro-2025-03-19", true},
+		{"O1-PRO uppercase", "O1-PRO", true},
+
 		// Other models - should NOT support reasoning
 		{"text-davinci-003", "text-davinci-003", false},
 		{"gpt-3", "gpt-3", false},
@@ -70,6 +80,44 @@ func TestIsOpenAIReasoningModel(t *testing.T) {
 	}
 }
 
+func TestSupportsReasoningSummary(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name     string
+		model    string
+		expected bool
+	}{
+		// Standard reasoning models - should support summary
+		{"o1-preview", "o1-preview", true},
+		{"o1-mini", "o1-mini", true},
+		{"o3-mini", "o3-mini", true},
+		{"o4-preview", "o4-preview", true},
+		{"gpt-5", "gpt-5", true},
+		{"gpt-5-mini", "gpt-5-mini", true},
+
+		// o1-pro models - do NOT support summary
+		{"o1-pro", "o1-pro", false},
+		{"o1-pro-2025-03-19", "o1-pro-2025-03-19", false},
+		{"O1-PRO uppercase", "O1-PRO", false},
+
+		// Non-reasoning models - do NOT support summary
+		{"gpt-4o", "gpt-4o", false},
+		{"gpt-4-turbo", "gpt-4-turbo", false},
+		{"gpt-5-chat-latest", "gpt-5-chat-latest", false},
+		{"empty string", "", false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			result := supportsReasoningSummary(tt.model)
+			assert.Equal(t, tt.expected, result, "Model %s should return %v", tt.model, tt.expected)
+		})
+	}
+}
+
 func TestGetOpenAIReasoningEffort_Success(t *testing.T) {
 	t.Parallel()