add GPT-5 model constants and reasoning validation (sashabaranov#1062)

Author: behzadsp

chat_test.go

@@ -334,6 +334,126 @@ func TestO3ModelsChatCompletionsBetaLimitations(t *testing.T) {
 	}
 }
 
+func TestGPT5ModelsChatCompletionsBetaLimitations(t *testing.T) {
+	tests := []struct {
+		name          string
+		in            openai.ChatCompletionRequest
+		expectedError error
+	}{
+		{
+			name: "log_probs_unsupported",
+			in: openai.ChatCompletionRequest{
+				MaxCompletionTokens: 1000,
+				LogProbs:            true,
+				Model:               openai.GPT5,
+			},
+			expectedError: openai.ErrReasoningModelLimitationsLogprobs,
+		},
+		{
+			name: "set_temperature_unsupported",
+			in: openai.ChatCompletionRequest{
+				MaxCompletionTokens: 1000,
+				Model:               openai.GPT5Mini,
+				Messages: []openai.ChatCompletionMessage{
+					{
+						Role: openai.ChatMessageRoleUser,
+					},
+					{
+						Role: openai.ChatMessageRoleAssistant,
+					},
+				},
+				Temperature: float32(2),
+			},
+			expectedError: openai.ErrReasoningModelLimitationsOther,
+		},
+		{
+			name: "set_top_unsupported",
+			in: openai.ChatCompletionRequest{
+				MaxCompletionTokens: 1000,
+				Model:               openai.GPT5Nano,
+				Messages: []openai.ChatCompletionMessage{
+					{
+						Role: openai.ChatMessageRoleUser,
+					},
+					{
+						Role: openai.ChatMessageRoleAssistant,
+					},
+				},
+				Temperature: float32(1),
+				TopP:        float32(0.1),
+			},
+			expectedError: openai.ErrReasoningModelLimitationsOther,
+		},
+		{
+			name: "set_n_unsupported",
+			in: openai.ChatCompletionRequest{
+				MaxCompletionTokens: 1000,
+				Model:               openai.GPT5ChatLatest,
+				Messages: []openai.ChatCompletionMessage{
+					{
+						Role: openai.ChatMessageRoleUser,
+					},
+					{
+						Role: openai.ChatMessageRoleAssistant,
+					},
+				},
+				Temperature: float32(1),
+				TopP:        float32(1),
+				N:           2,
+			},
+			expectedError: openai.ErrReasoningModelLimitationsOther,
+		},
+		{
+			name: "set_presence_penalty_unsupported",
+			in: openai.ChatCompletionRequest{
+				MaxCompletionTokens: 1000,
+				Model:               openai.GPT5,
+				Messages: []openai.ChatCompletionMessage{
+					{
+						Role: openai.ChatMessageRoleUser,
+					},
+					{
+						Role: openai.ChatMessageRoleAssistant,
+					},
+				},
+				PresencePenalty: float32(0.1),
+			},
+			expectedError: openai.ErrReasoningModelLimitationsOther,
+		},
+		{
+			name: "set_frequency_penalty_unsupported",
+			in: openai.ChatCompletionRequest{
+				MaxCompletionTokens: 1000,
+				Model:               openai.GPT5Mini,
+				Messages: []openai.ChatCompletionMessage{
+					{
+						Role: openai.ChatMessageRoleUser,
+					},
+					{
+						Role: openai.ChatMessageRoleAssistant,
+					},
+				},
+				FrequencyPenalty: float32(0.1),
+			},
+			expectedError: openai.ErrReasoningModelLimitationsOther,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			config := openai.DefaultConfig("whatever")
+			config.BaseURL = "http://localhost/v1"
+			client := openai.NewClientWithConfig(config)
+			ctx := context.Background()
+
+			_, err := client.CreateChatCompletion(ctx, tt.in)
+			checks.HasError(t, err)
+			msg := fmt.Sprintf("CreateChatCompletion should return wrong model error, returned: %s", err)
+			checks.ErrorIs(t, err, tt.expectedError, msg)
+		})
+	}
+}
+
 func TestChatRequestOmitEmpty(t *testing.T) {
 	data, err := json.Marshal(openai.ChatCompletionRequest{
 		// We set model b/c it's required, so omitempty doesn't make sense

completion.go

@@ -49,6 +49,10 @@ const (
 	GPT4Dot1Nano20250414    = "gpt-4.1-nano-2025-04-14"
 	GPT4Dot5Preview         = "gpt-4.5-preview"
 	GPT4Dot5Preview20250227 = "gpt-4.5-preview-2025-02-27"
+	GPT5                    = "gpt-5"
+	GPT5Mini                = "gpt-5-mini"
+	GPT5Nano                = "gpt-5-nano"
+	GPT5ChatLatest          = "gpt-5-chat-latest"
 	GPT3Dot5Turbo0125       = "gpt-3.5-turbo-0125"
 	GPT3Dot5Turbo1106       = "gpt-3.5-turbo-1106"
 	GPT3Dot5Turbo0613       = "gpt-3.5-turbo-0613"
@@ -142,6 +146,10 @@ var disabledModelsForEndpoints = map[string]map[string]bool{
 		GPT4Dot1Mini20250414:    true,
 		GPT4Dot1Nano:            true,
 		GPT4Dot1Nano20250414:    true,
+		GPT5:                    true,
+		GPT5Mini:                true,
+		GPT5Nano:                true,
+		GPT5ChatLatest:          true,
 	},
 	chatCompletionsSuffix: {
 		CodexCodeDavinci002:     true,

completion_test.go

@@ -300,3 +300,32 @@ func TestCompletionWithGPT4oModels(t *testing.T) {
 		})
 	}
 }
+
+// TestCompletionWithGPT5Models Tests that GPT5 models are not supported for completion endpoint.
+func TestCompletionWithGPT5Models(t *testing.T) {
+	config := openai.DefaultConfig("whatever")
+	config.BaseURL = "http://localhost/v1"
+	client := openai.NewClientWithConfig(config)
+
+	models := []string{
+		openai.GPT5,
+		openai.GPT5Mini,
+		openai.GPT5Nano,
+		openai.GPT5ChatLatest,
+	}
+
+	for _, model := range models {
+		t.Run(model, func(t *testing.T) {
+			_, err := client.CreateCompletion(
+				context.Background(),
+				openai.CompletionRequest{
+					MaxTokens: 5,
+					Model:     model,
+				},
+			)
+			if !errors.Is(err, openai.ErrCompletionUnsupportedModel) {
+				t.Fatalf("CreateCompletion should return ErrCompletionUnsupportedModel for %s model, but returned: %v", model, err)
+			}
+		})
+	}
+}

reasoning_validator.go

@@ -28,21 +28,22 @@ var (
 	ErrReasoningModelLimitationsOther    = errors.New("this model has beta-limitations, temperature, top_p and n are fixed at 1, while presence_penalty and frequency_penalty are fixed at 0") //nolint:lll
 )
 
-// ReasoningValidator handles validation for o-series model requests.
+// ReasoningValidator handles validation for reasoning model requests.
 type ReasoningValidator struct{}
 
-// NewReasoningValidator creates a new validator for o-series models.
+// NewReasoningValidator creates a new validator for reasoning models.
 func NewReasoningValidator() *ReasoningValidator {
 	return &ReasoningValidator{}
 }
 
-// Validate performs all validation checks for o-series models.
+// Validate performs all validation checks for reasoning models.
 func (v *ReasoningValidator) Validate(request ChatCompletionRequest) error {
 	o1Series := strings.HasPrefix(request.Model, "o1")
 	o3Series := strings.HasPrefix(request.Model, "o3")
 	o4Series := strings.HasPrefix(request.Model, "o4")
+	gpt5Series := strings.HasPrefix(request.Model, "gpt-5")
 
-	if !o1Series && !o3Series && !o4Series {
+	if !o1Series && !o3Series && !o4Series && !gpt5Series {
 		return nil
 	}