feat: sync deployed v1.0.0-rc.38-custom.42

Sync production source including routing, billing, Astra detection and public uptime. Update custom build instructions and license packaging.

relaykit/dto/billing_usage.go

@@ -88,7 +88,8 @@ func HasOpenAIUsageTokens(usage *Usage) bool {
 		usage.ClaudeCacheCreation1hTokens != 0 {
 		return true
 	}
-	if usage.PromptTokensDetails.CachedTokens != 0 ||
+	if usage.PromptTokensDetails.CachedTokensDetails.HasTokens() ||
+		usage.PromptTokensDetails.CachedTokens != 0 ||
 		usage.PromptTokensDetails.CachedCreationTokens != 0 ||
 		usage.PromptTokensDetails.CacheWriteTokens != 0 ||
 		usage.PromptTokensDetails.TextTokens != 0 ||
@@ -105,7 +106,8 @@ func HasOpenAIUsageTokens(usage *Usage) bool {
 	if usage.InputTokensDetails == nil {
 		return false
 	}
-	return usage.InputTokensDetails.CachedTokens != 0 ||
+	return usage.InputTokensDetails.CachedTokensDetails.HasTokens() ||
+		usage.InputTokensDetails.CachedTokens != 0 ||
 		usage.InputTokensDetails.CachedCreationTokens != 0 ||
 		usage.InputTokensDetails.CacheWriteTokens != 0 ||
 		usage.InputTokensDetails.TextTokens != 0 ||
@@ -385,8 +387,9 @@ func cloneOpenAIUsage(usage *Usage) *Usage {
 	}
 	clone := *usage
 	clone.BillingUsage = nil
+	clone.PromptTokensDetails = usage.PromptTokensDetails.Clone()
 	if usage.InputTokensDetails != nil {
-		inputTokensDetails := *usage.InputTokensDetails
+		inputTokensDetails := usage.InputTokensDetails.Clone()
 		clone.InputTokensDetails = &inputTokensDetails
 	}
 	return &clone

relaykit/dto/billing_usage_test.go

@@ -49,6 +49,33 @@ func TestNewOpenAIChatBillingUsageRequiresTokenContent(t *testing.T) {
 	assert.Equal(t, 1, billingUsage.OpenAIUsage.PromptTokens)
 }
 
+func TestImageCacheDetailsSurviveUsageSnapshots(t *testing.T) {
+	var original Usage
+	require.NoError(t, kitutil.Unmarshal([]byte(`{"input_tokens":1000,"input_tokens_details":{"cached_tokens":300,"image_tokens":600,"cached_tokens_details":{"image_tokens":200,"text_tokens":100}}}`), &original))
+	billing := NewOpenAIResponsesBillingUsage(&original)
+	require.NotNil(t, billing)
+	*original.InputTokensDetails.CachedTokensDetails.ImageTokens = 99
+	canonical, ok := billing.CanonicalUsage()
+	require.True(t, ok)
+	require.NotNil(t, canonical.PromptTokensDetails.CachedTokensDetails)
+	assert.Equal(t, 200, *canonical.PromptTokensDetails.CachedTokensDetails.ImageTokens)
+	*canonical.PromptTokensDetails.CachedTokensDetails.ImageTokens = 42
+	assert.Equal(t, 200, *billing.OpenAIUsage.InputTokensDetails.CachedTokensDetails.ImageTokens)
+
+	var incoming Usage
+	require.NoError(t, kitutil.Unmarshal([]byte(`{"input_tokens_details":{"cached_tokens_details":{"image_tokens":0}}}`), &incoming))
+	merged := MergeUsageNonZero(canonical, &incoming)
+	require.NotNil(t, merged.InputTokensDetails.CachedTokensDetails.ImageTokens)
+	assert.Zero(t, *merged.InputTokensDetails.CachedTokensDetails.ImageTokens)
+	assert.Equal(t, 100, *merged.InputTokensDetails.CachedTokensDetails.TextTokens)
+	*incoming.InputTokensDetails.CachedTokensDetails.ImageTokens = 9
+	assert.Zero(t, *merged.InputTokensDetails.CachedTokensDetails.ImageTokens)
+	encoded, err := kitutil.Marshal(merged)
+	require.NoError(t, err)
+	assert.Contains(t, string(encoded), `"cached_tokens_details":{"text_tokens":100,"image_tokens":0}`)
+	assert.NotContains(t, string(encoded), `"audio_tokens":null`)
+}
+
 func TestNewEstimatedGeminiChatBillingUsage(t *testing.T) {
 	billingUsage := NewEstimatedGeminiChatBillingUsage(&Usage{
 		PromptTokens:     11,

relaykit/dto/channel_settings.go

@@ -11,13 +11,16 @@ import (
 )
 
 type ChannelSettings struct {
-	TaskPluginKey          string `json:"task_plugin_key,omitempty"`
-	ForceFormat            bool   `json:"force_format,omitempty"`
-	ThinkingToContent      bool   `json:"thinking_to_content,omitempty"`
-	Proxy                  string `json:"proxy"`
-	PassThroughBodyEnabled bool   `json:"pass_through_body_enabled,omitempty"`
-	SystemPrompt           string `json:"system_prompt,omitempty"`
-	SystemPromptOverride   bool   `json:"system_prompt_override,omitempty"`
+	TaskPluginKey             string `json:"task_plugin_key,omitempty"`
+	ForceFormat               bool   `json:"force_format,omitempty"`
+	ThinkingToContent         bool   `json:"thinking_to_content,omitempty"`
+	Proxy                     string `json:"proxy"`
+	PassThroughBodyEnabled    bool   `json:"pass_through_body_enabled,omitempty"`
+	ResponsesWebSocketEnabled bool   `json:"responses_websocket_enabled,omitempty"`
+	// Deprecated: retained for stored settings compatibility; tokens control the guard.
+	ResponsesAstraModelGuard *bool  `json:"responses_astra_model_guard,omitempty"`
+	SystemPrompt             string `json:"system_prompt,omitempty"`
+	SystemPromptOverride     bool   `json:"system_prompt_override,omitempty"`
 	// HTTPProtocol controls outbound HTTP version negotiation for this channel.
 	// Accepted values: "", "auto" (default), "http1".
 	HTTPProtocol string `json:"http_protocol,omitempty"`
@@ -26,6 +29,10 @@ type ChannelSettings struct {
 	HTTP2ConnectionShards int `json:"http2_connection_shards,omitempty"`
 }
 
+func (s ChannelSettings) AstraModelGuardEnabled() bool {
+	return s.ResponsesAstraModelGuard == nil || *s.ResponsesAstraModelGuard
+}
+
 const (
 	HTTPProtocolAuto         = "auto"
 	HTTPProtocolHTTP1        = "http1"
@@ -86,6 +93,10 @@ type ChannelOtherSettings struct {
 	UpstreamModelUpdateLastRemovedModels  []string              `json:"upstream_model_update_last_removed_models,omitempty"`  // 上次检测到的可删除模型
 	UpstreamModelUpdateIgnoredModels      []string              `json:"upstream_model_update_ignored_models,omitempty"`       // 手动忽略的模型
 	AdvancedCustom                        *AdvancedCustomConfig `json:"advanced_custom,omitempty"`
+	// OllamaOpenAIChat routes Ollama chat completions to the OpenAI-compatible
+	// /v1/chat/completions endpoint. When unset, chat completions keep using
+	// the native /api/chat protocol.
+	OllamaOpenAIChat bool `json:"ollama_openai_chat,omitempty"`
 	// ToolLossPolicy is a channel-level opt-in for request-phase conversion
 	// rejection. Empty follows the default allow policy. Accepted values:
 	// "", "allow", "safe", "strict".
@@ -114,6 +125,7 @@ func (s *ChannelOtherSettings) ValidateToolLossPolicy() error {
 }
 
 const (
+	AdvancedCustomConverterSGLangRerank                = "jina_rerank_to_sglang"
 	advancedCustomConverterNone                        = "none"
 	advancedCustomConverterClaudeMessagesToOpenAIChat  = "anthropic_messages_to_openai_chat_completions"
 	advancedCustomConverterOpenAIChatToClaudeMessages  = "openai_chat_completions_to_anthropic_messages"
@@ -135,11 +147,22 @@ type AdvancedCustomConfig struct {
 }
 
 type AdvancedCustomRoute struct {
-	IncomingPath string                   `json:"incoming_path,omitempty"`
-	UpstreamPath string                   `json:"upstream_path,omitempty"`
-	Converter    string                   `json:"converter,omitempty"`
-	Models       []string                 `json:"models,omitempty"`
-	Auth         *AdvancedCustomRouteAuth `json:"auth,omitempty"`
+	IncomingPath           string                   `json:"incoming_path,omitempty"`
+	UpstreamPath           string                   `json:"upstream_path,omitempty"`
+	Converter              string                   `json:"converter,omitempty"`
+	Models                 []string                 `json:"models,omitempty"`
+	Auth                   *AdvancedCustomRouteAuth `json:"auth,omitempty"`
+	PassThroughBodyEnabled bool                     `json:"pass_through_body_enabled,omitempty"`
+}
+
+// SupportsPassThroughBody reports whether the route converter leaves the request body untouched.
+func (r AdvancedCustomRoute) SupportsPassThroughBody() bool {
+	switch strings.TrimSpace(r.Converter) {
+	case "", advancedCustomConverterNone, AdvancedCustomConverterSGLangRerank:
+		return true
+	default:
+		return false
+	}
 }
 
 type AdvancedCustomRouteAuth struct {
@@ -171,6 +194,14 @@ const (
 	AdvancedCustomBalancePath = "/v1/dashboard/billing/credit_grants"
 )
 
+// IsNative reports whether the route forwards requests without protocol
+// conversion. Validate normalizes an empty converter to none, but callers may
+// see configurations that were never saved.
+func (r AdvancedCustomRoute) IsNative() bool {
+	converter := strings.TrimSpace(r.Converter)
+	return converter == "" || converter == advancedCustomConverterNone
+}
+
 // MatchPath returns the first route whose IncomingPath matches requestPath.
 // Matching mirrors the relay adaptor: exact match, {model} placeholder, and
 // :generateContent <-> :streamGenerateContent equivalence.
@@ -373,6 +404,7 @@ func matchAdvancedCustomIncomingPathTemplate(configuredPath string, requestPath
 func IsAdvancedCustomConverterAllowed(converter string) bool {
 	switch converter {
 	case advancedCustomConverterNone,
+		AdvancedCustomConverterSGLangRerank,
 		advancedCustomConverterClaudeMessagesToOpenAIChat,
 		advancedCustomConverterOpenAIChatToClaudeMessages,
 		advancedCustomConverterOpenAIChatToOpenAIResponses,
@@ -438,6 +470,9 @@ func (c *AdvancedCustomConfig) Validate() error {
 			if strings.Contains(upstreamPath, advancedCustomModelPlaceholder) {
 				return fmt.Errorf("advanced_custom.advanced_routes[%d].upstream_path must not contain %s for %s", i, advancedCustomModelPlaceholder, managementRouteName)
 			}
+			if route.PassThroughBodyEnabled {
+				return fmt.Errorf("advanced_custom.advanced_routes[%d].pass_through_body_enabled must be false for %s", i, managementRouteName)
+			}
 		}
 		if err := validateAdvancedCustomRouteModels(i, route.IncomingPath, route.Models, paths); err != nil {
 			return err
@@ -456,6 +491,9 @@ func (c *AdvancedCustomConfig) Validate() error {
 		if err := validateAdvancedCustomConverterPath(i, route.IncomingPath, route.Converter); err != nil {
 			return err
 		}
+		if route.PassThroughBodyEnabled && !route.SupportsPassThroughBody() {
+			return fmt.Errorf("advanced_custom.advanced_routes[%d].pass_through_body_enabled requires converter none: %s", i, route.Converter)
+		}
 		if err := validateAdvancedCustomRouteAuth(i, route.Auth); err != nil {
 			return err
 		}
@@ -556,6 +594,9 @@ func validateAdvancedCustomUpstreamTarget(index int, upstreamPath string) error
 }
 
 func validateAdvancedCustomConverterPath(index int, incomingPath string, converter string) error {
+	if converter == AdvancedCustomConverterSGLangRerank && (incomingPath == "/v1/rerank" || incomingPath == "/rerank") {
+		return nil
+	}
 	if incomingPath == advancedCustomEndpointPathOpenAIAlphaSearch {
 		if converter == advancedCustomConverterNone {
 			return nil

relaykit/dto/channel_settings_test.go

@@ -585,6 +585,84 @@ func TestAdvancedCustomValidateAlphaSearchConverterPath(t *testing.T) {
 	}
 }
 
+func TestAdvancedCustomValidateRoutePassThroughBody(t *testing.T) {
+	valid := &AdvancedCustomConfig{
+		Routes: []AdvancedCustomRoute{
+			{
+				IncomingPath:           "/v1/chat/completions",
+				UpstreamPath:           "/v1/chat/completions",
+				PassThroughBodyEnabled: true,
+			},
+			{
+				IncomingPath:           "/v1/rerank",
+				UpstreamPath:           "/v1/rerank",
+				Converter:              AdvancedCustomConverterSGLangRerank,
+				PassThroughBodyEnabled: true,
+			},
+			{
+				IncomingPath: "/v1/messages",
+				UpstreamPath: "/v1/chat/completions",
+				Converter:    advancedCustomConverterClaudeMessagesToOpenAIChat,
+			},
+		},
+	}
+	require.NoError(t, valid.Validate())
+
+	route, ok := valid.MatchPathForModel("/v1/chat/completions", "gpt-4o")
+	require.True(t, ok)
+	assert.True(t, route.PassThroughBodyEnabled)
+	route, ok = valid.MatchPathForModel("/v1/messages", "gpt-4o")
+	require.True(t, ok)
+	assert.False(t, route.PassThroughBodyEnabled)
+
+	encoded, err := json.Marshal(valid.Routes[2])
+	require.NoError(t, err)
+	assert.NotContains(t, string(encoded), "pass_through_body_enabled")
+
+	tests := []struct {
+		name  string
+		route AdvancedCustomRoute
+		want  string
+	}{
+		{
+			name: "converter route",
+			route: AdvancedCustomRoute{
+				IncomingPath:           "/v1/messages",
+				UpstreamPath:           "/v1/chat/completions",
+				Converter:              advancedCustomConverterClaudeMessagesToOpenAIChat,
+				PassThroughBodyEnabled: true,
+			},
+			want: "pass_through_body_enabled requires converter none",
+		},
+		{
+			name: "model list route",
+			route: AdvancedCustomRoute{
+				IncomingPath:           AdvancedCustomModelListPath,
+				UpstreamPath:           "/v1/models",
+				PassThroughBodyEnabled: true,
+			},
+			want: "pass_through_body_enabled must be false for /v1/models",
+		},
+		{
+			name: "balance route",
+			route: AdvancedCustomRoute{
+				IncomingPath:           AdvancedCustomBalancePath,
+				UpstreamPath:           "/provider/balance",
+				PassThroughBodyEnabled: true,
+			},
+			want: "pass_through_body_enabled must be false for /v1/dashboard/billing/credit_grants",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			err := (&AdvancedCustomConfig{Routes: []AdvancedCustomRoute{tt.route}}).Validate()
+			require.Error(t, err)
+			assert.Contains(t, err.Error(), tt.want)
+		})
+	}
+}
+
 func TestChannelSettingsHTTPTransportJSONRoundTrip(t *testing.T) {
 	legacy := `{"proxy":"http://127.0.0.1:8080","force_format":true}`
 	var settings ChannelSettings

relaykit/dto/claude.go

@@ -246,6 +246,8 @@ type ClaudeRequest struct {
 	Thinking          *Thinking       `json:"thinking,omitempty"`
 	McpServers        json.RawMessage `json:"mcp_servers,omitempty"`
 	Metadata          json.RawMessage `json:"metadata,omitempty"`
+	// vLLM Messages extension; forwarded verbatim when present.
+	ChatTemplateKwargs json.RawMessage `json:"chat_template_kwargs,omitempty"`
 	// Speed specifies the Claude inference speed mode.
 	// This field is filtered by default and can be enabled via channel setting allow_speed.
 	Speed json.RawMessage `json:"speed,omitempty"`

relaykit/dto/legacy_dalle_image.go

@@ -0,0 +1,58 @@
+package dto
+
+// Deprecated model compatibility: OpenAI removed DALL·E 2 and DALL·E 3 from
+// its API on 2026-05-12. Retain these rules for existing configurations and
+// compatible upstreams; do not extend them for current image models.
+// https://developers.openai.com/api/docs/deprecations#2025-11-14-dalle-model-snapshots
+
+import (
+	"errors"
+	"strings"
+)
+
+// NormalizeLegacyDalleImageRequest preserves the retired models' request
+// validation and defaults. Other models, including aliases, are unchanged.
+func (i *ImageRequest) NormalizeLegacyDalleImageRequest() error {
+	switch i.Model {
+	case "dall-e-2", "dall-e":
+		if i.Size != "" && i.Size != "256x256" && i.Size != "512x512" && i.Size != "1024x1024" {
+			return errors.New("size must be one of 256x256, 512x512, or 1024x1024 for dall-e-2 or dall-e")
+		}
+		if i.Size == "" {
+			i.Size = "1024x1024"
+		}
+	case "dall-e-3":
+		if i.Size != "" && i.Size != "1024x1024" && i.Size != "1024x1792" && i.Size != "1792x1024" {
+			return errors.New("size must be one of 1024x1024, 1024x1792 or 1792x1024 for dall-e-3")
+		}
+		if i.Quality == "" {
+			i.Quality = "standard"
+		}
+		if i.Size == "" {
+			i.Size = "1024x1024"
+		}
+	}
+	return nil
+}
+
+func (i *ImageRequest) legacyDallePriceRatio() float64 {
+	if !strings.HasPrefix(i.Model, "dall-e") {
+		return 1
+	}
+	sizeRatio, qualityRatio := 1.0, 1.0
+	switch i.Size {
+	case "256x256":
+		sizeRatio = 0.4
+	case "512x512":
+		sizeRatio = 0.45
+	case "1024x1792", "1792x1024":
+		sizeRatio = 2
+	}
+	if i.Model == "dall-e-3" && i.Quality == "hd" {
+		qualityRatio = 2
+		if i.Size == "1024x1792" || i.Size == "1792x1024" {
+			qualityRatio = 1.5
+		}
+	}
+	return sizeRatio * qualityRatio
+}

relaykit/dto/openai_compaction.go

@@ -9,7 +9,7 @@ import (
 type OpenAIResponsesCompactionResponse struct {
 	ID        string          `json:"id"`
 	Object    string          `json:"object"`
-	CreatedAt int             `json:"created_at"`
+	CreatedAt IntValue        `json:"created_at"`
 	Output    json.RawMessage `json:"output"`
 	Usage     *Usage          `json:"usage"`
 	Error     any             `json:"error,omitempty"`

relaykit/dto/openai_image.go

@@ -2,9 +2,9 @@ package dto
 
 import (
 	"encoding/json"
+	"fmt"
 	"net/http"
 	"reflect"
-	"strings"
 
 	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
 	"github.com/QuantumNous/new-api/relaykit/types"
@@ -14,6 +14,13 @@ import (
 // wrapped-negative n overflows quota calculation into a negative charge.
 const MaxImageN = 128
 
+// ImageBillingParameters contains only the provider scalars parsed by request
+// validation. Keep this separate from the complete provider request payload.
+type ImageBillingParameters struct {
+	N            *uint `json:"n,omitempty"`
+	PromptExtend *bool `json:"prompt_extend,omitempty"`
+}
+
 type ImageRequest struct {
 	Model             string          `json:"model"`
 	Prompt            string          `json:"prompt" binding:"required"`
@@ -39,7 +46,29 @@ type ImageRequest struct {
 	UserId           json.RawMessage `json:"user_id,omitempty"`
 	Image            json.RawMessage `json:"image,omitempty"`
 	// 用匿名参数接收额外参数
-	Extra map[string]json.RawMessage `json:"-"`
+	Extra             map[string]json.RawMessage `json:"-"`
+	BillingParameters *ImageBillingParameters    `json:"-"`
+}
+
+// ImageCount resolves the validated request quantity. Top-level zero retains
+// its legacy default of one; an explicit provider count must be positive.
+func (i *ImageRequest) ImageCount(useProviderParameters bool) (int, error) {
+	n := uint(1)
+	if i.N != nil && *i.N != 0 {
+		n = *i.N
+	}
+	if n > MaxImageN {
+		return 0, fmt.Errorf("n must be an integer between 1 and %d", MaxImageN)
+	}
+	if parameters := i.BillingParameters; parameters != nil && parameters.N != nil {
+		if *parameters.N > MaxImageN || useProviderParameters && *parameters.N == 0 {
+			return 0, fmt.Errorf("parameters.n must be an integer between 1 and %d", MaxImageN)
+		}
+		if useProviderParameters {
+			n = *parameters.N
+		}
+	}
+	return int(n), nil
 }
 
 func (i *ImageRequest) UnmarshalJSON(data []byte) error {
@@ -131,29 +160,6 @@ func indexComma(s string) int {
 }
 
 func (i *ImageRequest) GetTokenCountMeta() *types.TokenCountMeta {
-	var sizeRatio = 1.0
-	var qualityRatio = 1.0
-
-	if strings.HasPrefix(i.Model, "dall-e") {
-		// Size
-		if i.Size == "256x256" {
-			sizeRatio = 0.4
-		} else if i.Size == "512x512" {
-			sizeRatio = 0.45
-		} else if i.Size == "1024x1024" {
-			sizeRatio = 1
-		} else if i.Size == "1024x1792" || i.Size == "1792x1024" {
-			sizeRatio = 2
-		}
-
-		if i.Model == "dall-e-3" && i.Quality == "hd" {
-			qualityRatio = 2.0
-			if i.Size == "1024x1792" || i.Size == "1792x1024" {
-				qualityRatio = 1.5
-			}
-		}
-	}
-
 	imageN := uint(1)
 	if i.N != nil && *i.N > 0 {
 		imageN = *i.N
@@ -165,7 +171,7 @@ func (i *ImageRequest) GetTokenCountMeta() *types.TokenCountMeta {
 	return &types.TokenCountMeta{
 		CombineText:     i.Prompt,
 		MaxTokens:       1584,
-		ImagePriceRatio: sizeRatio * qualityRatio,
+		ImagePriceRatio: i.legacyDallePriceRatio(),
 		BillingRatios:   map[string]float64{"n": float64(imageN)},
 	}
 }

relaykit/dto/openai_request.go

@@ -109,6 +109,25 @@ type GeneralOpenAIRequest struct {
 	ReasoningSplit json.RawMessage `json:"reasoning_split,omitempty"`
 	// vLLM
 	ThinkingTokenBudget json.RawMessage `json:"thinking_token_budget,omitempty"`
+	IncludeReasoning    json.RawMessage `json:"include_reasoning,omitempty"`
+	MinP                json.RawMessage `json:"min_p,omitempty"`
+	RepetitionPenalty   json.RawMessage `json:"repetition_penalty,omitempty"`
+	StructuredOutputs   json.RawMessage `json:"structured_outputs,omitempty"`
+	ReturnTokenIds      json.RawMessage `json:"return_token_ids,omitempty"`
+	// SGLang OpenAI-compatible sampling and reasoning controls (v0.5.19).
+	// Native /generate sampling_params and server routing controls are not chat fields.
+	MinTokens            *uint           `json:"min_tokens,omitempty"`
+	SeparateReasoning    json.RawMessage `json:"separate_reasoning,omitempty"`
+	StreamReasoning      json.RawMessage `json:"stream_reasoning,omitempty"`
+	Regex                json.RawMessage `json:"regex,omitempty"`
+	EBNF                 json.RawMessage `json:"ebnf,omitempty"`
+	StopTokenIDs         json.RawMessage `json:"stop_token_ids,omitempty"`
+	StopRegex            json.RawMessage `json:"stop_regex,omitempty"`
+	NoStopTrim           json.RawMessage `json:"no_stop_trim,omitempty"`
+	IgnoreEOS            json.RawMessage `json:"ignore_eos,omitempty"`
+	SkipSpecialTokens    json.RawMessage `json:"skip_special_tokens,omitempty"`
+	ContinueFinalMessage json.RawMessage `json:"continue_final_message,omitempty"`
+	CacheSalt            json.RawMessage `json:"cache_salt,omitempty"`
 
 	// Internal conversion state; never serialized to an upstream protocol.
 	ReasoningConversion *ReasoningConversionState `json:"-"`
@@ -386,6 +405,9 @@ type StreamOptions struct {
 	// IncludeObfuscation is only for /v1/responses stream payload.
 	// This field is filtered by default and can be enabled via channel setting allow_include_obfuscation.
 	IncludeObfuscation bool `json:"include_obfuscation,omitempty"`
+	// ContinuousUsageStats is a vLLM stream_options extension that emits
+	// usage on intermediate chunks. Optional so an explicit false is kept.
+	// ContinuousUsageStats *bool `json:"continuous_usage_stats,omitempty"`
 }
 
 func (r *GeneralOpenAIRequest) GetMaxTokens() uint {
@@ -1029,6 +1051,14 @@ type OpenAIResponsesRequest struct {
 	// qwen
 	EnableThinking json.RawMessage `json:"enable_thinking,omitempty"`
 	ThinkingBudget json.RawMessage `json:"thinking_budget,omitempty"`
+	// vLLM
+	ChatTemplateKwargs json.RawMessage `json:"chat_template_kwargs,omitempty"`
+	// SGLang Responses sampling extensions.
+	TopK              json.RawMessage `json:"top_k,omitempty"`
+	MinP              json.RawMessage `json:"min_p,omitempty"`
+	RepetitionPenalty json.RawMessage `json:"repetition_penalty,omitempty"`
+	Stop              json.RawMessage `json:"stop,omitempty"`
+	CacheSalt         json.RawMessage `json:"cache_salt,omitempty"`
 	// perplexity
 	Preset json.RawMessage `json:"preset,omitempty"`
 

relaykit/dto/openai_request_zero_value_test.go

@@ -307,3 +307,23 @@ func TestGeneralOpenAIRequestPreserveMessageLevelTools(t *testing.T) {
 	assert.Contains(t, meta.CombineText, "Get the current time of a city")
 	assert.Contains(t, meta.CombineText, "lookup_order")
 }
+
+func TestSGLangRequestExtensionsRoundTrip(t *testing.T) {
+	for _, tc := range []struct {
+		name    string
+		raw     string
+		request any
+	}{
+		{"chat", `{"model":"served-model","min_tokens":0,"top_k":-1,"min_p":0,"repetition_penalty":1,"separate_reasoning":false,"stream_reasoning":false,"regex":"[0-9]+","ebnf":"root ::= \"yes\"","stop_token_ids":[0],"stop_regex":["END$"],"no_stop_trim":false,"ignore_eos":false,"skip_special_tokens":false,"continue_final_message":false,"cache_salt":"tenant","chat_template_kwargs":{"enable_thinking":false}}`, &GeneralOpenAIRequest{}},
+		{"responses", `{"model":"served-model","input":"hi","top_k":-1,"min_p":0,"repetition_penalty":1,"frequency_penalty":0,"presence_penalty":0,"stop":["END"],"cache_salt":"tenant","chat_template_kwargs":{"enable_thinking":false}}`, &OpenAIResponsesRequest{}},
+		{"chat omitted", `{"model":"served-model"}`, &GeneralOpenAIRequest{}},
+		{"responses omitted", `{"model":"served-model"}`, &OpenAIResponsesRequest{}},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			require.NoError(t, kitutil.Unmarshal([]byte(tc.raw), tc.request))
+			encoded, err := kitutil.Marshal(tc.request)
+			require.NoError(t, err)
+			assert.JSONEq(t, tc.raw, string(encoded))
+		})
+	}
+}

relaykit/dto/openai_response.go

@@ -234,11 +234,12 @@ type Usage struct {
 	UsageSource          string        `json:"usage_source,omitempty"`
 	BillingUsage         *BillingUsage `json:"billing_usage,omitempty"`
 
-	PromptTokensDetails    InputTokenDetails  `json:"prompt_tokens_details"`
-	CompletionTokenDetails OutputTokenDetails `json:"completion_tokens_details"`
-	InputTokens            int                `json:"input_tokens"`
-	OutputTokens           int                `json:"output_tokens"`
-	InputTokensDetails     *InputTokenDetails `json:"input_tokens_details"`
+	PromptTokensDetails    InputTokenDetails   `json:"prompt_tokens_details"`
+	CompletionTokenDetails OutputTokenDetails  `json:"completion_tokens_details"`
+	InputTokens            int                 `json:"input_tokens"`
+	OutputTokens           int                 `json:"output_tokens"`
+	InputTokensDetails     *InputTokenDetails  `json:"input_tokens_details"`
+	OutputTokensDetails    *OutputTokenDetails `json:"output_tokens_details,omitempty"`
 
 	// claude cache 1h
 	ClaudeCacheCreation5mTokens int `json:"claude_cache_creation_5_m_tokens"`
@@ -259,8 +260,9 @@ type OpenAIVideoResponse struct {
 }
 
 type InputTokenDetails struct {
-	CachedTokens         int `json:"cached_tokens"`
-	CachedCreationTokens int `json:"cached_creation_tokens,omitempty"`
+	CachedTokens         int                 `json:"cached_tokens"`
+	CachedTokensDetails  *CachedTokenDetails `json:"cached_tokens_details,omitempty"`
+	CachedCreationTokens int                 `json:"cached_creation_tokens,omitempty"`
 	// CacheWriteTokens is OpenAI's native cache-write count, reported as
 	// prompt_tokens_details.cache_write_tokens (Chat Completions) or
 	// input_tokens_details.cache_write_tokens (Responses). It is billed at the
@@ -271,6 +273,40 @@ type InputTokenDetails struct {
 	ImageTokens      int `json:"image_tokens"`
 }
 
+// CachedTokenDetails describes subsets of cached_tokens. Pointers distinguish
+// an unreported modality from an explicitly reported zero.
+type CachedTokenDetails struct {
+	TextTokens  *int `json:"text_tokens,omitempty"`
+	ImageTokens *int `json:"image_tokens,omitempty"`
+	AudioTokens *int `json:"audio_tokens,omitempty"`
+}
+
+func (d *CachedTokenDetails) HasTokens() bool {
+	return d != nil && (d.TextTokens != nil && *d.TextTokens != 0 ||
+		d.ImageTokens != nil && *d.ImageTokens != 0 || d.AudioTokens != nil && *d.AudioTokens != 0)
+}
+
+// Clone preserves presence information without sharing mutable usage fields.
+func (d InputTokenDetails) Clone() InputTokenDetails {
+	if d.CachedTokensDetails != nil {
+		cached := *d.CachedTokensDetails
+		if cached.TextTokens != nil {
+			value := *cached.TextTokens
+			cached.TextTokens = &value
+		}
+		if cached.ImageTokens != nil {
+			value := *cached.ImageTokens
+			cached.ImageTokens = &value
+		}
+		if cached.AudioTokens != nil {
+			value := *cached.AudioTokens
+			cached.AudioTokens = &value
+		}
+		d.CachedTokensDetails = &cached
+	}
+	return d
+}
+
 // CacheCreationTokensTotal returns the cache-write token count regardless of
 // which field the upstream reported it in: Claude-derived conversions populate
 // CachedCreationTokens while OpenAI reports cache_write_tokens natively. Both
@@ -295,7 +331,7 @@ type OutputTokenDetails struct {
 type OpenAIResponsesResponse struct {
 	ID                 string             `json:"id"`
 	Object             string             `json:"object"`
-	CreatedAt          int                `json:"created_at"`
+	CreatedAt          IntValue           `json:"created_at"`
 	Status             json.RawMessage    `json:"status"`
 	Error              any                `json:"error,omitempty"`
 	IncompleteDetails  *IncompleteDetails `json:"incomplete_details,omitempty"`

relaykit/dto/usage_merge.go

@@ -50,7 +50,7 @@ func MergeUsageNonZero(current *Usage, incoming *Usage) *Usage {
 			details.CacheWriteTokens > 0 ||
 			details.TextTokens > 0 ||
 			details.AudioTokens > 0 ||
-			details.ImageTokens > 0 {
+			details.ImageTokens > 0 || details.CachedTokensDetails != nil {
 			if current.InputTokensDetails == nil {
 				current.InputTokensDetails = &InputTokenDetails{}
 			}
@@ -286,6 +286,25 @@ func mergeGeminiTokenDetails(current []GeminiPromptTokensDetails, incoming []Gem
 }
 
 func mergeInputTokenDetails(current *InputTokenDetails, incoming InputTokenDetails) {
+	if incoming.CachedTokensDetails != nil {
+		// Unlike legacy scalar counters, these optional fields explicitly report
+		// zero. Merge only present modalities and detach the resulting snapshot.
+		merged := current.Clone()
+		if merged.CachedTokensDetails == nil {
+			merged.CachedTokensDetails = &CachedTokenDetails{}
+		}
+		details := incoming.Clone().CachedTokensDetails
+		if details.TextTokens != nil {
+			merged.CachedTokensDetails.TextTokens = details.TextTokens
+		}
+		if details.ImageTokens != nil {
+			merged.CachedTokensDetails.ImageTokens = details.ImageTokens
+		}
+		if details.AudioTokens != nil {
+			merged.CachedTokensDetails.AudioTokens = details.AudioTokens
+		}
+		current.CachedTokensDetails = merged.CachedTokensDetails
+	}
 	if incoming.CachedTokens > 0 {
 		current.CachedTokens = incoming.CachedTokens
 	}

relaykit/dto/user_settings.go

@@ -16,6 +16,7 @@ type UserSetting struct {
 	SidebarModules                   string  `json:"sidebar_modules,omitempty"`                      // SidebarModules 左侧边栏模块配置
 	BillingPreference                string  `json:"billing_preference,omitempty"`                   // BillingPreference 扣费策略（订阅/钱包）
 	Language                         string  `json:"language,omitempty"`                             // Language 用户语言偏好 (zh, en)
+	AstraCostNoticeAck               bool    `json:"astra_cost_notice_ack,omitempty"`                // Astra 费用说明已确认
 }
 
 var (

relaykit/dto/values.go

@@ -37,6 +37,11 @@ func (i *IntValue) UnmarshalJSON(b []byte) error {
 		*i = IntValue(n)
 		return nil
 	}
+	var f float64
+	if err := kitutil.Unmarshal(b, &f); err == nil {
+		*i = IntValue(int(f))
+		return nil
+	}
 	var s string
 	if err := kitutil.Unmarshal(b, &s); err != nil {
 		return err

relaykit/dto/values_test.go

@@ -0,0 +1,47 @@
+package dto
+
+import (
+	"testing"
+
+	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestIntValueUnmarshalJSON(t *testing.T) {
+	testCases := []struct {
+		name     string
+		input    string
+		expected int
+	}{
+		{name: "integer", input: `1768488160`, expected: 1768488160},
+		{name: "float", input: `1786588600.0`, expected: 1786588600},
+		{name: "scientific", input: `1.76848816E9`, expected: 1768488160},
+		{name: "string integer", input: `"42"`, expected: 42},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			var v IntValue
+			require.NoError(t, kitutil.UnmarshalJsonStr(tc.input, &v))
+			assert.Equal(t, tc.expected, int(v))
+		})
+	}
+}
+
+func TestResponsesStreamResponseCreatedAtFloat(t *testing.T) {
+	var resp ResponsesStreamResponse
+	require.NoError(t, kitutil.UnmarshalJsonStr(`{"response":{"id":"resp_1","created_at":1786588600.0}}`, &resp))
+	require.NotNil(t, resp.Response)
+	assert.Equal(t, 1786588600, int(resp.Response.CreatedAt))
+
+	encoded, err := kitutil.Marshal(resp.Response)
+	require.NoError(t, err)
+	assert.Contains(t, string(encoded), `"created_at":1786588600`)
+	assert.NotContains(t, string(encoded), `1786588600.0`)
+}
+
+func TestResponsesCompactionResponseCreatedAtFloat(t *testing.T) {
+	var resp OpenAIResponsesCompactionResponse
+	require.NoError(t, kitutil.UnmarshalJsonStr(`{"id":"resp_1","object":"response","created_at":1.76848816E9}`, &resp))
+	assert.Equal(t, 1768488160, int(resp.CreatedAt))
+}