Jaymr/new-api
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: bill OpenAI cache_write_tokens at cache-creation price with zero clamp

Browse Source 
Parse OpenAI's native cache_write_tokens (chat prompt_tokens_details /
responses input_tokens_details), bill it at the cache-creation ratio, and
clamp the uncached prompt remainder at zero since cached + cache-write can
exceed prompt_tokens. Propagate the field through chat/responses/claude
format conversions and tiered expression billing (cc variable).
This commit is contained in:
CaIon
2026-07-11 21:18:49 +08:00
parent c36418c863
commit 48068ce923
18 changed files with 158 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
dto/billing_usage.go
+1
View File
@@ -88,6 +88,7 @@ func HasOpenAIUsageTokens(usage *Usage) bool {
}
if usage.PromptTokensDetails.CachedTokens != 0 ||
usage.PromptTokensDetails.CachedCreationTokens != 0 ||
usage.PromptTokensDetails.CacheWriteTokens != 0 ||
usage.PromptTokensDetails.TextTokens != 0 ||
usage.PromptTokensDetails.ImageTokens != 0 ||
usage.PromptTokensDetails.AudioTokens != 0 {
dto/openai_request.go
+2
View File
@@ -849,6 +849,7 @@ type OpenAIResponsesRequest struct {
MaxOutputTokens *uint `json:"max_output_tokens,omitempty"`
TopLogProbs *int `json:"top_logprobs,omitempty"`
Metadata json.RawMessage `json:"metadata,omitempty"`
Moderation json.RawMessage `json:"moderation,omitempty"`
ParallelToolCalls json.RawMessage `json:"parallel_tool_calls,omitempty"`
PreviousResponseID string `json:"previous_response_id,omitempty"`
Reasoning *Reasoning `json:"reasoning,omitempty"`
@@ -859,6 +860,7 @@ type OpenAIResponsesRequest struct {
// This field is allowed by default and can be disabled via channel setting disable_store.
Store json.RawMessage `json:"store,omitempty"`
PromptCacheKey json.RawMessage `json:"prompt_cache_key,omitempty"`
PromptCacheOptions json.RawMessage `json:"prompt_cache_options,omitempty"`
PromptCacheRetention json.RawMessage `json:"prompt_cache_retention,omitempty"`
// SafetyIdentifier carries client identity for policy abuse detection.
// This field is filtered by default and can be enabled via channel setting allow_safety_identifier.
dto/openai_response.go
+20 -3
View File
@@ -256,9 +256,26 @@ type OpenAIVideoResponse struct {
type InputTokenDetails struct {
CachedTokens int `json:"cached_tokens"`
CachedCreationTokens int `json:"cached_creation_tokens,omitempty"`
TextTokens int `json:"text_tokens"`
AudioTokens int `json:"audio_tokens"`
ImageTokens int `json:"image_tokens"`
// 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
// cache-creation price.
CacheWriteTokens int `json:"cache_write_tokens,omitempty"`
TextTokens int `json:"text_tokens"`
AudioTokens int `json:"audio_tokens"`
ImageTokens int `json:"image_tokens"`
}
// 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
// are billed at the cache-creation price; when both are present the larger
// value wins so the same tokens are never double-counted.
func (d InputTokenDetails) CacheCreationTokensTotal() int {
if d.CacheWriteTokens > d.CachedCreationTokens {
return d.CacheWriteTokens
}
return d.CachedCreationTokens
}
type OutputTokenDetails struct {
dto/openai_responses_compaction_request.go
+8 -6
View File
@@ -17,12 +17,14 @@ type OpenAIResponsesCompactionRequest struct {
// Codex compact request parity:
// https://github.com/openai/codex/commit/53d59722268dde82fb93c1f37964ce196c2a86d7
// https://github.com/openai/codex/commit/5d6f23a27bf9c90709af527a7108c1c2eadf5123
Tools json.RawMessage `json:"tools,omitempty"`
ParallelToolCalls json.RawMessage `json:"parallel_tool_calls,omitempty"`
Reasoning *Reasoning `json:"reasoning,omitempty"`
ServiceTier string `json:"service_tier,omitempty"`
PromptCacheKey json.RawMessage `json:"prompt_cache_key,omitempty"`
Text json.RawMessage `json:"text,omitempty"`
Tools json.RawMessage `json:"tools,omitempty"`
ParallelToolCalls json.RawMessage `json:"parallel_tool_calls,omitempty"`
Reasoning *Reasoning `json:"reasoning,omitempty"`
ServiceTier string `json:"service_tier,omitempty"`
PromptCacheKey json.RawMessage `json:"prompt_cache_key,omitempty"`
PromptCacheOptions json.RawMessage `json:"prompt_cache_options,omitempty"`
PromptCacheRetention json.RawMessage `json:"prompt_cache_retention,omitempty"`
Text json.RawMessage `json:"text,omitempty"`
}
func (r *OpenAIResponsesCompactionRequest) GetTokenCountMeta() *types.TokenCountMeta {
relay/channel/openai/relay_responses.go
+2
View File
@@ -51,6 +51,7 @@ func OaiResponsesHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http
usage.TotalTokens = responsesResponse.Usage.TotalTokens
if responsesResponse.Usage.InputTokensDetails != nil {
usage.PromptTokensDetails.CachedTokens = responsesResponse.Usage.InputTokensDetails.CachedTokens
usage.PromptTokensDetails.CacheWriteTokens = responsesResponse.Usage.InputTokensDetails.CacheWriteTokens
}
}
if info == nil || info.ResponsesUsageInfo == nil || info.ResponsesUsageInfo.BuiltInTools == nil {
@@ -104,6 +105,7 @@ func OaiResponsesStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp
}
if streamResponse.Response.Usage.InputTokensDetails != nil {
usage.PromptTokensDetails.CachedTokens = streamResponse.Response.Usage.InputTokensDetails.CachedTokens
usage.PromptTokensDetails.CacheWriteTokens = streamResponse.Response.Usage.InputTokensDetails.CacheWriteTokens
}
}
if streamResponse.Response.HasImageGenerationCall() {
relay/channel/openai/relay_responses_compact.go
+1
View File
@@ -37,6 +37,7 @@ func OaiResponsesCompactionHandler(c *gin.Context, resp *http.Response) (*dto.Us
usage.TotalTokens = compactResp.Usage.TotalTokens
if compactResp.Usage.InputTokensDetails != nil {
usage.PromptTokensDetails.CachedTokens = compactResp.Usage.InputTokensDetails.CachedTokens
usage.PromptTokensDetails.CacheWriteTokens = compactResp.Usage.InputTokensDetails.CacheWriteTokens
}
}
relay/responses_handler.go
+8 -6
View File
@@ -41,12 +41,14 @@ func ResponsesHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *
responsesReq = req
case *dto.OpenAIResponsesCompactionRequest:
responsesReq = &dto.OpenAIResponsesRequest{
Model: req.Model,
Input: req.Input,
Instructions: req.Instructions,
PreviousResponseID: req.PreviousResponseID,
ParallelToolCalls: req.ParallelToolCalls,
ServiceTier: req.ServiceTier,
Model: req.Model,
Input: req.Input,
Instructions: req.Instructions,
PreviousResponseID: req.PreviousResponseID,
ParallelToolCalls: req.ParallelToolCalls,
ServiceTier: req.ServiceTier,
PromptCacheOptions: req.PromptCacheOptions,
PromptCacheRetention: req.PromptCacheRetention,
}
default:
return types.NewErrorWithStatusCode(
service/relayconvert/internal/claude_messages/to_oai_chat_resp.go
+3
View File
@@ -216,6 +216,9 @@ func buildOpenAIStyleUsageFromClaudeUsage(usage *dto.Usage) dto.Usage {
usage.ClaudeCacheCreation1hTokens,
)
cacheCreationTokens := cacheCreationTokensForOpenAIUsage(usage)
// Expose the standard OpenAI cache-write field alongside the legacy
// cached_creation_tokens so OpenAI-format clients can bill cache writes.
clone.PromptTokensDetails.CacheWriteTokens = cacheCreationTokens
totalInputTokens := usage.PromptTokens + usage.PromptTokensDetails.CachedTokens + cacheCreationTokens
clone.PromptTokens = totalInputTokens
clone.InputTokens = totalInputTokens
service/relayconvert/internal/oai_chat/to_claude_messages_resp.go
+13 -2
View File
@@ -39,10 +39,21 @@ func buildClaudeUsageFromOpenAIUsage(oaiUsage *dto.Usage) *dto.ClaudeUsage {
oaiUsage.ClaudeCacheCreation5mTokens,
oaiUsage.ClaudeCacheCreation1hTokens,
)
cacheCreationTokens := oaiUsage.PromptTokensDetails.CacheCreationTokensTotal()
inputTokens := oaiUsage.PromptTokens
if oaiUsage.PromptTokensDetails.CacheWriteTokens > 0 {
// OpenAI native cache-write usage counts cached and cache-write tokens
// inside prompt_tokens, while Claude semantics reports input_tokens
// excluding both; the uncached remainder clamps at zero.
inputTokens = oaiUsage.PromptTokens - oaiUsage.PromptTokensDetails.CachedTokens - cacheCreationTokens
if inputTokens < 0 {
inputTokens = 0
}
}
usage := &dto.ClaudeUsage{
InputTokens: oaiUsage.PromptTokens,
InputTokens: inputTokens,
OutputTokens: oaiUsage.CompletionTokens,
CacheCreationInputTokens: oaiUsage.PromptTokensDetails.CachedCreationTokens,
CacheCreationInputTokens: cacheCreationTokens,
CacheReadInputTokens: oaiUsage.PromptTokensDetails.CachedTokens,
BillingUsage: billingUsage,
}
service/relayconvert/internal/oai_chat/to_claude_messages_resp_test.go
+24
View File
@@ -79,6 +79,30 @@ func TestResponseOpenAI2ClaudeUsageCarriesOpenAIBillingUsage(t *testing.T) {
assert.Nil(t, resp.Usage.BillingUsage.OpenAIUsage.BillingUsage)
}
func TestBuildClaudeUsageFromOpenAICacheWriteUsage(t *testing.T) {
usage := buildClaudeUsageFromOpenAIUsage(&dto.Usage{
PromptTokens: 3619,
CompletionTokens: 36,
TotalTokens: 3655,
PromptTokensDetails: dto.InputTokenDetails{
CachedTokens: 2921,
CacheWriteTokens: 3616,
},
})
require.NotNil(t, usage)
// Claude semantics reports input_tokens excluding cache read/write; the
// remainder 3619-2921-3616 clamps to 0.
assert.Equal(t, 0, usage.InputTokens)
assert.Equal(t, 2921, usage.CacheReadInputTokens)
assert.Equal(t, 3616, usage.CacheCreationInputTokens)
assert.Equal(t, 36, usage.OutputTokens)
require.NotNil(t, usage.BillingUsage)
require.NotNil(t, usage.BillingUsage.OpenAIUsage)
assert.Equal(t, dto.BillingUsageSemanticOpenAI, usage.BillingUsage.Semantic)
assert.Equal(t, 3616, usage.BillingUsage.OpenAIUsage.PromptTokensDetails.CacheWriteTokens)
}
func TestStreamResponseOpenAI2ClaudeClosesTextThinkingAndToolBlocks(t *testing.T) {
info := &relaycommon.RelayInfo{
ClaudeConvertInfo: &relaycommon.ClaudeConvertInfo{
service/relayconvert/internal/oai_chat/to_oai_responses_resp.go
+1
View File
@@ -137,6 +137,7 @@ func UsageFromChatUsage(src *dto.Usage) *dto.Usage {
src.PromptTokensDetails.ImageTokens != 0 ||
src.PromptTokensDetails.AudioTokens != 0 ||
src.PromptTokensDetails.CachedCreationTokens != 0 ||
src.PromptTokensDetails.CacheWriteTokens != 0 ||
src.PromptTokensDetails.TextTokens != 0 {
details := src.PromptTokensDetails
usage.InputTokensDetails = &details
service/relayconvert/internal/oai_responses/to_oai_chat_resp.go
+1
View File
@@ -156,6 +156,7 @@ func UsageFromResponsesUsage(src *dto.Usage) *dto.Usage {
if src.InputTokensDetails != nil {
usage.PromptTokensDetails.CachedTokens = src.InputTokensDetails.CachedTokens
usage.PromptTokensDetails.CachedCreationTokens = src.InputTokensDetails.CachedCreationTokens
usage.PromptTokensDetails.CacheWriteTokens = src.InputTokensDetails.CacheWriteTokens
usage.PromptTokensDetails.TextTokens = src.InputTokensDetails.TextTokens
usage.PromptTokensDetails.ImageTokens = src.InputTokensDetails.ImageTokens
usage.PromptTokensDetails.AudioTokens = src.InputTokensDetails.AudioTokens
service/relayconvert/response_registry_test.go
+5
View File
@@ -277,6 +277,7 @@ func TestConvertResponseProviderToOAIChatUsage(t *testing.T) {
assert.Equal(t, 22, toChat.Usage.TotalTokens)
assert.Equal(t, 3, toChat.Usage.PromptTokensDetails.CachedTokens)
assert.Equal(t, 4, toChat.Usage.PromptTokensDetails.CachedCreationTokens)
assert.Equal(t, 4, toChat.Usage.PromptTokensDetails.CacheWriteTokens)
require.NotNil(t, toChat.Usage.BillingUsage)
require.NotNil(t, toChat.Usage.BillingUsage.ClaudeUsage)
assert.Equal(t, dto.BillingUsageSourceClaudeMessages, toChat.Usage.BillingUsage.Source)
@@ -539,6 +540,7 @@ func TestResponseUsageMatrixChatAndResponsesDetails(t *testing.T) {
PromptTokensDetails: dto.InputTokenDetails{
CachedTokens: 3,
CachedCreationTokens: 2,
CacheWriteTokens: 6,
TextTokens: 4,
AudioTokens: 1,
ImageTokens: 5,
@@ -558,6 +560,7 @@ func TestResponseUsageMatrixChatAndResponsesDetails(t *testing.T) {
require.NotNil(t, result.Usage.InputTokensDetails)
assert.Equal(t, 3, result.Usage.InputTokensDetails.CachedTokens)
assert.Equal(t, 2, result.Usage.InputTokensDetails.CachedCreationTokens)
assert.Equal(t, 6, result.Usage.InputTokensDetails.CacheWriteTokens)
assert.Equal(t, 4, result.Usage.InputTokensDetails.TextTokens)
assert.Equal(t, 1, result.Usage.InputTokensDetails.AudioTokens)
assert.Equal(t, 5, result.Usage.InputTokensDetails.ImageTokens)
@@ -579,6 +582,7 @@ func TestResponseUsageMatrixChatAndResponsesDetails(t *testing.T) {
InputTokensDetails: &dto.InputTokenDetails{
CachedTokens: 4,
CachedCreationTokens: 1,
CacheWriteTokens: 7,
TextTokens: 5,
AudioTokens: 2,
ImageTokens: 1,
@@ -598,6 +602,7 @@ func TestResponseUsageMatrixChatAndResponsesDetails(t *testing.T) {
assert.Equal(t, 21, result.Usage.TotalTokens)
assert.Equal(t, 4, result.Usage.PromptTokensDetails.CachedTokens)
assert.Equal(t, 1, result.Usage.PromptTokensDetails.CachedCreationTokens)
assert.Equal(t, 7, result.Usage.PromptTokensDetails.CacheWriteTokens)
assert.Equal(t, 5, result.Usage.PromptTokensDetails.TextTokens)
assert.Equal(t, 2, result.Usage.PromptTokensDetails.AudioTokens)
assert.Equal(t, 1, result.Usage.PromptTokensDetails.ImageTokens)
service/text_quota.go
+8 -1
View File
@@ -208,7 +208,7 @@ func calculateTextQuotaSummary(ctx *gin.Context, relayInfo *relaycommon.RelayInf
summary.CompletionTokens = usage.CompletionTokens
summary.TotalTokens = usage.PromptTokens + usage.CompletionTokens
summary.CacheTokens = usage.PromptTokensDetails.CachedTokens
summary.CacheCreationTokens = usage.PromptTokensDetails.CachedCreationTokens
summary.CacheCreationTokens = usage.PromptTokensDetails.CacheCreationTokensTotal()
summary.CacheCreationTokens5m = usage.ClaudeCacheCreation5mTokens
summary.CacheCreationTokens1h = usage.ClaudeCacheCreation1hTokens
summary.ImageTokens = usage.PromptTokensDetails.ImageTokens
@@ -294,6 +294,13 @@ func calculateTextQuotaSummary(ctx *gin.Context, relayInfo *relaycommon.RelayInf
}
}
// OpenAI cache-write usage can report cached_tokens + cache_write_tokens
// exceeding prompt_tokens; the uncached remainder must clamp at zero so
// billing never subtracts more than the reported input.
if baseTokens.IsNegative() {
baseTokens = decimal.Zero
}
promptQuota := baseTokens.Add(cachedTokensWithRatio).Add(imageTokensWithRatio).Add(cachedCreationTokensWithRatio)
completionQuota := dCompletionTokens.Mul(dCompletionRatio)
quotaCalculateDecimal := promptQuota.Add(completionQuota).Mul(ratio)
service/text_quota_test.go
+56
View File
@@ -375,6 +375,62 @@ func TestCalculateTextQuotaSummaryHandlesLegacyClaudeDerivedOpenAIUsage(t *testi
require.Equal(t, 1624, summary.Quota)
}
func TestCalculateTextQuotaSummaryBillsOpenAICacheWriteTokens(t *testing.T) {
gin.SetMode(gin.TestMode)
w := httptest.NewRecorder()
ctx, _ := gin.CreateTestContext(w)
relayInfo := &relaycommon.RelayInfo{
RelayFormat: types.RelayFormatOpenAI,
OriginModelName: "gpt-5.1",
PriceData: types.PriceData{
ModelRatio: 1,
CompletionRatio: 2,
CacheRatio: 0.1,
CacheCreationRatio: 1.25,
GroupRatioInfo: types.GroupRatioInfo{GroupRatio: 1},
},
StartTime: time.Now(),
}
t.Run("uncached remainder stays positive", func(t *testing.T) {
usage := &dto.Usage{
PromptTokens: 1473,
CompletionTokens: 19,
PromptTokensDetails: dto.InputTokenDetails{
CacheWriteTokens: 1470,
},
}
summary := calculateTextQuotaSummary(ctx, relayInfo, usage)
require.Equal(t, 1470, summary.CacheCreationTokens)
// (1473-0-1470) + 1470*1.25 + 19*2 = 3 + 1837.5 + 38 = 1878.5 => 1879
require.Equal(t, 1879, summary.Quota)
})
t.Run("uncached remainder clamps to zero", func(t *testing.T) {
// Real OpenAI payload shape: cached_tokens + cache_write_tokens exceeds
// prompt_tokens, so the uncached remainder must clamp to 0 instead of
// producing a negative charge component.
usage := &dto.Usage{
PromptTokens: 3619,
CompletionTokens: 36,
PromptTokensDetails: dto.InputTokenDetails{
CachedTokens: 2921,
CacheWriteTokens: 3616,
},
}
summary := calculateTextQuotaSummary(ctx, relayInfo, usage)
require.Equal(t, 3619, summary.PromptTokens)
require.Equal(t, 3616, summary.CacheCreationTokens)
// max(3619-2921-3616, 0) + 2921*0.1 + 3616*1.25 + 36*2 = 4884.1 => 4884
require.Equal(t, 4884, summary.Quota)
})
}
func TestCalculateTextQuotaSummarySeparatesOpenRouterCacheReadFromPromptBilling(t *testing.T) {
gin.SetMode(gin.TestMode)
w := httptest.NewRecorder()
service/tiered_settle.go
+1 -1
View File
@@ -22,7 +22,7 @@ func BuildTieredTokenParams(usage *dto.Usage, isClaudeUsageSemantic bool, usedVa
p := float64(usage.PromptTokens)
c := float64(usage.CompletionTokens)
cr := float64(usage.PromptTokensDetails.CachedTokens)
cc5m := float64(usage.PromptTokensDetails.CachedCreationTokens)
cc5m := float64(usage.PromptTokensDetails.CacheCreationTokensTotal())
cc1h := float64(0)
if usage.UsageSemantic == "anthropic" {
setting/ratio_setting/cache_ratio.go
+3
View File
@@ -81,6 +81,9 @@ var defaultCacheRatio = map[string]float64{
}
var defaultCreateCacheRatio = map[string]float64{
"gpt-5.6-sol": 1.25,
"gpt-5.6-terra": 1.25,
"gpt-5.6-luna": 1.25,
"claude-3-sonnet-20240229": 1.25,
"claude-3-opus-20240229": 1.25,
"claude-3-haiku-20240307": 1.25,
web/default/src/features/usage-logs/components/usage-logs-mobile-card.tsx
Vendored
+1 -1
View File
@@ -329,7 +329,7 @@ function CommonLogsCard<TData>({
/>
</div>
<div className='grid grid-cols-[minmax(0,1.35fr)_minmax(0,0.75fr)] gap-1.5'>
<div className='grid grid-cols-[minmax(0,1fr)_minmax(0,0.8fr)] gap-1.5'>
<div className='bg-muted/20 min-w-0 rounded-md px-2 py-1.5'>
<MobileLogTimeStatus
createdAt={rowData?.created_at}