feat(relay): hosted-tool conversion fidelity, reasoning normalization, and billing usage integrity (#7137)

* feat(relaykit): preserve hosted tools across conversions

- add protocol-neutral hosted-tool DTOs, conversion metadata, and loss policies
- bridge citations, grounding metadata, and hosted-tool stream lifecycles
- document the public conversion behavior and channel policy controls

* refactor(relaykit): normalize reasoning and thinking intent

- centralize provider-neutral reasoning intent, effort, and budget mappings
- parse model suffixes at the host entry boundary while preserving provider-owned tails
- keep adaptive Claude thinking and explicit zero-token compatibility consistent

* fix(billing): preserve authoritative usage across relay hops

- carry native BillingUsage sidecars through direct and streamed protocol bridges
- merge partial and terminal usage monotonically with safe fallback settlement
- retain cache metadata, penultimate usage, and per-call Gemini tool surcharges

* feat(relay): bridge Responses with Claude and Gemini protocols

- add direct request, response, and stream converters across supported relay formats
- expose Claude count_tokens and Chat-to-Responses compatibility endpoints
- carry conversion diagnostics through the host while retaining the curated public goldens

* fix(relay): wire relaykit conversions into host channels

- connect handlers, adaptors, and channel settings to the standalone conversion layer
- keep model mapping, pricing identity, retries, and provider-specific suffix behavior aligned
- ignore local audit artifacts and retain focused public regression coverage

setting/model_setting/global.go

@@ -35,6 +35,7 @@ func (p ChatCompletionsToResponsesPolicy) IsChannelEnabled(channelID int, channe
 type GlobalSettings struct {
 	PassThroughRequestEnabled        bool                             `json:"pass_through_request_enabled"`
 	ThinkingModelBlacklist           []string                         `json:"thinking_model_blacklist"`
+	EffortTailModelIDs               []string                         `json:"effort_tail_model_ids"`
 	ChatCompletionsToResponsesPolicy ChatCompletionsToResponsesPolicy `json:"chat_completions_to_responses_policy"`
 }
 
@@ -45,6 +46,13 @@ var defaultOpenaiSettings = GlobalSettings{
 		"moonshotai/kimi-k2-thinking",
 		"kimi-k2-thinking",
 	},
+	EffortTailModelIDs: []string{
+		"gpt-5.1-codex-max",
+		"qwen-image-edit-max",
+		"qwen-max",
+		"stable-diffusion-3-medium",
+		"yi-medium",
+	},
 	ChatCompletionsToResponsesPolicy: ChatCompletionsToResponsesPolicy{
 		Enabled:     false,
 		AllChannels: true,
@@ -77,3 +85,26 @@ func ShouldPreserveThinkingSuffix(modelName string) bool {
 	}
 	return false
 }
+
+// ShouldPreserveEffortTail reports model IDs whose names already end in an
+// effort-like token and must not be treated as reasoning aliases.
+func ShouldPreserveEffortTail(modelName string) bool {
+	target := strings.TrimSpace(modelName)
+	if target == "" {
+		return false
+	}
+	bare := target
+	if slash := strings.LastIndex(bare, "/"); slash >= 0 {
+		bare = bare[slash+1:]
+	}
+	for _, entry := range globalSettings.EffortTailModelIDs {
+		entry = strings.TrimSpace(entry)
+		if entry == "" {
+			continue
+		}
+		if entry == target || entry == bare {
+			return true
+		}
+	}
+	return false
+}

setting/ratio_setting/model_ratio.go

@@ -179,12 +179,10 @@ var defaultModelRatio = map[string]float64{
 	"gemini-2.5-pro-exp-03-25":                  0.625,
 	"gemini-2.5-pro-preview-03-25":              0.625,
 	"gemini-2.5-pro":                            0.625,
-	"gemini-2.5-flash-preview-04-17":            0.075,
-	"gemini-2.5-flash-preview-04-17-thinking":   0.075,
-	"gemini-2.5-flash-preview-04-17-nothinking": 0.075,
-	"gemini-2.5-flash-preview-05-20":            0.075,
-	"gemini-2.5-flash-preview-05-20-thinking":   0.075,
-	"gemini-2.5-flash-preview-05-20-nothinking": 0.075,
+	"gemini-2.5-flash-preview-04-17":          0.075,
+	"gemini-2.5-flash-preview-04-17-thinking": 0.075,
+	"gemini-2.5-flash-preview-05-20":          0.075,
+	"gemini-2.5-flash-preview-05-20-thinking": 0.075,
 	"gemini-2.5-flash-thinking-*":               0.075, // 用于为后续所有2.5 flash thinking budget 模型设置默认倍率
 	"gemini-2.5-pro-thinking-*":                 0.625, // 用于为后续所有2.5 pro thinking budget 模型设置默认倍率
 	"gemini-2.5-flash-lite-preview-thinking-*":  0.05,
@@ -549,9 +547,6 @@ func getHardcodedCompletionModelRatio(name string) (float64, bool) {
 			return 8, false
 		} else if strings.HasPrefix(name, "gemini-2.5-flash") { // 处理不同的flash模型倍率
 			if strings.HasPrefix(name, "gemini-2.5-flash-preview") {
-				if strings.HasSuffix(name, "-nothinking") {
-					return 4, false
-				}
 				return 3.5 / 0.15, false
 			}
 			if strings.HasPrefix(name, "gemini-2.5-flash-lite") {

setting/reasoning/suffix.go

@@ -1,9 +1,12 @@
 // Package reasoning re-exports the pure model-name effort-suffix helpers,
-// which moved to the conversion kit (service/relayconvert/reasoning) as part
+// which moved to the conversion kit (relaykit/relayconvert/reasoning) as part
 // of the relaykit extraction. Host code keeps importing this path unchanged.
 package reasoning
 
-import kitreasoning "github.com/QuantumNous/new-api/relaykit/relayconvert/reasoning"
+import (
+	kitreasoning "github.com/QuantumNous/new-api/relaykit/relayconvert/reasoning"
+	"github.com/QuantumNous/new-api/setting/model_setting"
+)
 
 var (
 	EffortSuffixes           = kitreasoning.EffortSuffixes
@@ -12,8 +15,13 @@ var (
 )
 
 var (
-	TrimEffortSuffix                          = kitreasoning.TrimEffortSuffix
-	TrimEffortSuffixWithSuffixes              = kitreasoning.TrimEffortSuffixWithSuffixes
-	ParseOpenAIReasoningEffortFromModelSuffix = kitreasoning.ParseOpenAIReasoningEffortFromModelSuffix
-	ParseDeepSeekV4ThinkingSuffix             = kitreasoning.ParseDeepSeekV4ThinkingSuffix
+	TrimEffortSuffixWithSuffixes  = kitreasoning.TrimEffortSuffixWithSuffixes
+	ParseDeepSeekV4ThinkingSuffix = kitreasoning.ParseDeepSeekV4ThinkingSuffix
+	TrimGeminiThinkingSuffix      = kitreasoning.TrimGeminiThinkingSuffix
 )
+
+// ParseOpenAIReasoningEffortFromModelSuffix applies the host effort-tail
+// whitelist so real model IDs such as qwen-max are not treated as aliases.
+func ParseOpenAIReasoningEffortFromModelSuffix(modelName string) (string, string) {
+	return kitreasoning.ParseOpenAIReasoningEffortFromModelSuffix(modelName, model_setting.ShouldPreserveEffortTail)
+}