Merge pull request #4994 from router-for-me/models

feat(models): add max_completion_tokens to model definitions and responses

internal/client/codex/models/models.go

@@ -61,6 +61,7 @@ func buildCodexClientModels(models []map[string]any, providersForModel Providers
 			entry := cloneCodexClientModelMap(template)
 			applyCodexClientDisplayName(entry, model)
 			applyCodexClientMaxContextLengthOverride(entry, model)
+			applyCodexClientMaxTokens(entry, model)
 			applyCodexClientSearchToolSupport(entry, id, true, providersForModel)
 			sanitizeCodexClientReasoningMetadata(entry)
 			applyCodexClientVisibilityOverride(entry, id)
@@ -73,6 +74,7 @@ func buildCodexClientModels(models []map[string]any, providersForModel Providers
 
 		entry := cloneCodexClientModelMap(defaultTemplate)
 		applyCodexClientModelMetadata(entry, id, model, optimizeMultiAgentV2)
+		applyCodexClientMaxTokens(entry, model)
 		applyCodexClientSearchToolSupport(entry, id, false, providersForModel)
 		sanitizeCodexClientReasoningMetadata(entry)
 		applyCodexClientVisibilityOverride(entry, id)
@@ -193,6 +195,12 @@ func applyCodexClientMaxContextLengthOverride(entry map[string]any, model map[st
 	}
 }
 
+func applyCodexClientMaxTokens(entry map[string]any, model map[string]any) {
+	if maxCompletionTokens := intModelValue(model, "max_completion_tokens"); maxCompletionTokens > 0 {
+		entry["max_tokens"] = maxCompletionTokens
+	}
+}
+
 func applyCodexClientSearchToolSupport(entry map[string]any, id string, templateModel bool, providersForModel ProvidersForModelFunc) {
 	supportsSearch, _ := entry["supports_search_tool"].(bool)
 	if !supportsSearch {

internal/client/codex/models/models_test.go

@@ -363,3 +363,38 @@ func TestCodexClientModelsResponseAppliesMaxContextLengthOverride(t *testing.T)
 		}
 	}
 }
+
+func TestCodexClientModelsResponseMapsMaxCompletionTokensToMaxTokens(t *testing.T) {
+	const wantTemplateLimit = 64000
+	const wantSynthesizedLimit = 32000
+
+	resp := BuildResponse([]map[string]any{
+		{"id": "gpt-5.5", "max_completion_tokens": wantTemplateLimit},
+		{"id": "custom-output-limit-model", "max_completion_tokens": wantSynthesizedLimit},
+	}, nil, false)
+	models, ok := resp["models"].([]map[string]any)
+	if !ok {
+		t.Fatalf("models type = %T, want []map[string]any", resp["models"])
+	}
+
+	bySlug := make(map[string]map[string]any, len(models))
+	for _, model := range models {
+		bySlug[stringModelValue(model, "slug")] = model
+	}
+
+	for _, testCase := range []struct {
+		slug string
+		want int
+	}{
+		{slug: "gpt-5.5", want: wantTemplateLimit},
+		{slug: "custom-output-limit-model", want: wantSynthesizedLimit},
+	} {
+		entry := bySlug[testCase.slug]
+		if entry == nil {
+			t.Fatalf("missing model %q", testCase.slug)
+		}
+		if got := intModelValue(entry, "max_tokens"); got != testCase.want {
+			t.Errorf("%s max_tokens = %d, want %d", testCase.slug, got, testCase.want)
+		}
+	}
+}