fix(relay): apply model-specific OpenAI chat capabilities (#7211)

* fix(relay): treat gpt-5 and later generations alike for max_completion_tokens

IsOpenAIGPT5Model matched on the literal prefix "gpt-5", so gpt-6-astra
(and every generation after it) fell through the gpt-5 request rules:
max_tokens was forwarded as-is and the provider rejected it with
"Unsupported parameter: 'max_tokens' is not supported with this model.
Use 'max_completion_tokens' instead." The same gap left temperature,
top_p and logprobs untouched, each of which the provider also rejects,
and made the channel test button report a 400 for a healthy deployment.

Match on the major version instead (gpt-<n>... with n >= 5). Callers are
unchanged: ConvertOpenAIRequest, GetSystemRoleName, buildTestRequest and
the health check all go through this one helper. buildTestRequest now
sends max_completion_tokens for these models directly instead of relying
on the later conversion. gpt-4.1, gpt-4o, gpt-oss, gpt-image and
gpt-realtime names still do not match.

Verified against Azure OpenAI gpt-6-astra (2026-09-03): with the old
prefix max_tokens / temperature / top_p / logprobs each returned 400,
while gpt-5.6-luna with the same payload returned 200.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016bwz8o5UeoRtrtDusKaayp

* fix(relay): separate OpenAI chat model compatibility rules

---------

Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: CaIon <i@caion.me>

relay/channel/openai/adaptor.go

@@ -357,34 +357,6 @@ func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayIn
 		info.SetReasoningEffort(string(effectiveEffort))
 		info.SetReasoningEffort(string(effectiveEffort))
 
 
 	}
 	}
-	isOModel := dto.IsOpenAIReasoningOModel(info.UpstreamModelName)
-	isGPT5Model := dto.IsOpenAIGPT5Model(info.UpstreamModelName)
-	if isOModel || isGPT5Model {
-		if lo.FromPtrOr(request.MaxCompletionTokens, uint(0)) == 0 && lo.FromPtrOr(request.MaxTokens, uint(0)) != 0 {
-			request.MaxCompletionTokens = request.MaxTokens
-			request.MaxTokens = nil
-		}
-
-		if isOModel {
-			request.Temperature = nil
-		}
-
-		// gpt-5系列模型适配 归零不再支持的参数
-		if isGPT5Model {
-			request.Temperature = nil
-			request.TopP = nil
-			request.LogProbs = nil
-		}
-
-		// o系列模型developer适配（o1-mini除外）
-		if !strings.HasPrefix(info.UpstreamModelName, "o1-mini") && !strings.HasPrefix(info.UpstreamModelName, "o1-preview") {
-			//修改第一个Message的内容，将system改为developer
-			if len(request.Messages) > 0 && request.Messages[0].Role == "system" {
-				request.Messages[0].Role = "developer"
-			}
-		}
-	}
-
 	if info.ChannelType != constant.ChannelTypeOpenRouter && renderReasoning {
 	if info.ChannelType != constant.ChannelTypeOpenRouter && renderReasoning {
 		effort, baseModel := reasoning.ParseOpenAIReasoningEffortFromModelSuffix(info.UpstreamModelName)
 		effort, baseModel := reasoning.ParseOpenAIReasoningEffortFromModelSuffix(info.UpstreamModelName)
 		if preserveSuffix {
 		if preserveSuffix {
@@ -431,6 +403,27 @@ func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayIn
 		}
 		}
 	}
 	}
 
 
+	capabilities := dto.GetOpenAIChatCapabilities(info.UpstreamModelName, info.ReasoningEffort)
+	if capabilities.UseMaxCompletionTokens {
+		if lo.FromPtrOr(request.MaxCompletionTokens, uint(0)) == 0 && lo.FromPtrOr(request.MaxTokens, uint(0)) != 0 {
+			request.MaxCompletionTokens = request.MaxTokens
+			request.MaxTokens = nil
+		}
+	}
+	if !capabilities.SupportsTemperature {
+		request.Temperature = nil
+	}
+	if !capabilities.SupportsTopP {
+		request.TopP = nil
+	}
+	if !capabilities.SupportsLogProbs {
+		request.LogProbs = nil
+		request.TopLogProbs = nil
+	}
+	if capabilities.UseDeveloperRole && len(request.Messages) > 0 && request.Messages[0].Role == "system" {
+		request.Messages[0].Role = "developer"
+	}
+
 	return request, nil
 	return request, nil
 }
 }
 
 

relaykit/dto/openai_request.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"fmt"
 	"net/http"
 	"net/http"
 	"strings"
 	"strings"
+	"time"
 
 
 	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
 	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
 	"github.com/QuantumNous/new-api/relaykit/types"
 	"github.com/QuantumNous/new-api/relaykit/types"
@@ -232,8 +233,74 @@ func IsOpenAIReasoningOModel(modelName string) bool {
 		strings.HasPrefix(modelName, "o4")
 		strings.HasPrefix(modelName, "o4")
 }
 }
 
 
+// IsOpenAIGPT5Model identifies the GPT-5 family, independently of request capabilities.
 func IsOpenAIGPT5Model(modelName string) bool {
 func IsOpenAIGPT5Model(modelName string) bool {
-	return strings.HasPrefix(modelName, "gpt-5")
+	return modelName == "gpt-5" || strings.HasPrefix(modelName, "gpt-5-") || strings.HasPrefix(modelName, "gpt-5.")
+}
+
+// OpenAIChatCapabilities describes independent Chat Completions compatibility rules.
+type OpenAIChatCapabilities struct {
+	UseMaxCompletionTokens bool
+	UseDeveloperRole       bool
+	SupportsTemperature    bool
+	SupportsTopP           bool
+	SupportsLogProbs       bool // Also governs top_logprobs.
+}
+
+// GetOpenAIChatCapabilities uses the mapped model and resolved reasoning effort.
+// Unrecognized models retain their parameters; future GPT generations do not
+// automatically inherit the restrictions of existing models.
+func GetOpenAIChatCapabilities(modelName, reasoningEffort string) OpenAIChatCapabilities {
+	capabilities := OpenAIChatCapabilities{
+		SupportsTemperature: true,
+		SupportsTopP:        true,
+		SupportsLogProbs:    true,
+	}
+	if IsOpenAIReasoningOModel(modelName) {
+		capabilities.UseMaxCompletionTokens = true
+		capabilities.UseDeveloperRole = !strings.HasPrefix(modelName, "o1-mini") && !strings.HasPrefix(modelName, "o1-preview")
+		capabilities.SupportsTemperature = false
+		return capabilities
+	}
+
+	isGPT5Model := IsOpenAIGPT5Model(modelName)
+	if !isGPT5Model && !isOpenAIModelSnapshot(modelName, "gpt-6-astra") {
+		return capabilities
+	}
+	capabilities.UseMaxCompletionTokens = true
+	capabilities.UseDeveloperRole = true
+
+	// These standard GPT-5 models default to none and support sampling only
+	// without reasoning. Named variants (pro, codex, chat-latest, etc.) do not
+	// inherit this exception. GPT-6 Astra never supports these parameters.
+	// https://developers.openai.com/api/docs/guides/latest-model?model=gpt-5.2
+	// https://developers.openai.com/api/docs/guides/latest-model?model=gpt-5.4
+	// https://developers.openai.com/api/docs/guides/latest-model?model=gpt-6-astra
+	supportsSampling := false
+	if isGPT5Model && (reasoningEffort == "" || reasoningEffort == "none") {
+		for _, model := range []string{"gpt-5.1", "gpt-5.2", "gpt-5.4"} {
+			if isOpenAIModelSnapshot(modelName, model) {
+				supportsSampling = true
+				break
+			}
+		}
+	}
+	capabilities.SupportsTemperature = supportsSampling
+	capabilities.SupportsTopP = supportsSampling
+	capabilities.SupportsLogProbs = supportsSampling
+	return capabilities
+}
+
+func isOpenAIModelSnapshot(modelName, baseModel string) bool {
+	if modelName == baseModel {
+		return true
+	}
+	snapshot, ok := strings.CutPrefix(modelName, baseModel+"-")
+	if !ok {
+		return false
+	}
+	_, err := time.Parse(time.DateOnly, snapshot)
+	return err == nil
 }
 }
 
 
 func IsQwenThinkingBudgetModel(modelName string) bool {
 func IsQwenThinkingBudgetModel(modelName string) bool {
@@ -245,11 +312,7 @@ func IsQwenThinkingBudgetModel(modelName string) bool {
 }
 }
 
 
 func (r *GeneralOpenAIRequest) GetSystemRoleName() string {
 func (r *GeneralOpenAIRequest) GetSystemRoleName() string {
-	if IsOpenAIReasoningOModel(r.Model) {
-		if !strings.HasPrefix(r.Model, "o1-mini") && !strings.HasPrefix(r.Model, "o1-preview") {
-			return "developer"
-		}
-	} else if IsOpenAIGPT5Model(r.Model) {
+	if GetOpenAIChatCapabilities(r.Model, r.ReasoningEffort).UseDeveloperRole {
 		return "developer"
 		return "developer"
 	}
 	}
 	return "system"
 	return "system"

relaykit/dto/openai_request_zero_value_test.go

@@ -195,6 +195,13 @@ func TestGeneralOpenAIRequestGetSystemRoleName(t *testing.T) {
 		{name: "o1 mini stays system", model: "o1-mini", want: "system"},
 		{name: "o1 mini stays system", model: "o1-mini", want: "system"},
 		{name: "o1 preview stays system", model: "o1-preview", want: "system"},
 		{name: "o1 preview stays system", model: "o1-preview", want: "system"},
 		{name: "gpt 5 uses developer", model: "gpt-5", want: "developer"},
 		{name: "gpt 5 uses developer", model: "gpt-5", want: "developer"},
+		{name: "gpt 5.6 uses developer", model: "gpt-5.6-luna", want: "developer"},
+		{name: "gpt 6 uses developer", model: "gpt-6-astra", want: "developer"},
+		{name: "gpt 6 snapshot uses developer", model: "gpt-6-astra-2026-09-03", want: "developer"},
+		{name: "unknown gpt 6 variant stays system", model: "gpt-6-astra-pro", want: "system"},
+		{name: "invalid gpt 6 snapshot stays system", model: "gpt-6-astra-2026-99-03", want: "system"},
+		{name: "unknown generation stays system", model: "gpt-7", want: "system"},
+		{name: "gpt 4.1 stays system", model: "gpt-4.1-nano", want: "system"},
 		{name: "omni is not o series", model: "omni-moderation-latest", want: "system"},
 		{name: "omni is not o series", model: "omni-moderation-latest", want: "system"},
 	}
 	}
 
 
@@ -202,7 +209,42 @@ func TestGeneralOpenAIRequestGetSystemRoleName(t *testing.T) {
 		t.Run(tt.name, func(t *testing.T) {
 		t.Run(tt.name, func(t *testing.T) {
 			req := GeneralOpenAIRequest{Model: tt.model}
 			req := GeneralOpenAIRequest{Model: tt.model}
 
 
-			require.Equal(t, tt.want, req.GetSystemRoleName())
+			assert.Equal(t, tt.want, req.GetSystemRoleName())
+		})
+	}
+}
+
+func TestIsOpenAIGPT5Model(t *testing.T) {
+	tests := []struct {
+		model string
+		want  bool
+	}{
+		{model: "gpt-5", want: true},
+		{model: "gpt-5-mini", want: true},
+		{model: "gpt-5-chat-latest", want: true},
+		{model: "gpt-5.6-luna", want: true},
+		{model: "gpt-5.4-nano", want: true},
+		{model: "gpt-5.2-2025-12-11", want: true},
+		{model: "gpt-6-astra", want: false},
+		{model: "gpt-50", want: false},
+		{model: "gpt-5custom", want: false},
+		{model: " GPT-5 ", want: false},
+		{model: "gpt-4.1", want: false},
+		{model: "gpt-4.1-nano", want: false},
+		{model: "gpt-4o", want: false},
+		{model: "gpt-4.5-preview", want: false},
+		{model: "gpt-oss-120b", want: false},
+		{model: "gpt-image-2", want: false},
+		{model: "gpt-realtime-2.1", want: false},
+		{model: "chatgpt-4o-latest", want: false},
+		{model: "o3-mini", want: false},
+		{model: "gpt-", want: false},
+		{model: "", want: false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.model, func(t *testing.T) {
+			assert.Equal(t, tt.want, IsOpenAIGPT5Model(tt.model))
 		})
 		})
 	}
 	}
 }
 }