wip: zen

cloud/app/src/routes/zen/v1/chat/completions.ts

@@ -1,8 +1,26 @@
 import type { APIEvent } from "@solidjs/start/server"
 import { handler } from "~/util/zen"
 
+type Usage = {
+  prompt_tokens?: number
+  completion_tokens?: number
+  total_tokens?: number
+  prompt_tokens_details?: {
+    text_tokens?: number
+    audio_tokens?: number
+    image_tokens?: number
+    cached_tokens?: number
+  }
+  completion_tokens_details?: {
+    reasoning_tokens?: number
+    audio_tokens?: number
+    accepted_prediction_tokens?: number
+    rejected_prediction_tokens?: number
+  }
+}
+
 export function POST(input: APIEvent) {
-  let usage: any
+  let usage: Usage
   return handler(input, {
     modifyBody: (body: any) => ({
       ...body,
@@ -17,7 +35,7 @@ export function POST(input: APIEvent) {
 
       let json
       try {
-        json = JSON.parse(chunk.slice(6))
+        json = JSON.parse(chunk.slice(6)) as { usage?: Usage }
       } catch (e) {
         return
       }
@@ -26,11 +44,11 @@ export function POST(input: APIEvent) {
       usage = json.usage
     },
     getStreamUsage: () => usage,
-    normalizeUsage: (usage: any) => ({
+    normalizeUsage: (usage: Usage) => ({
       inputTokens: usage.prompt_tokens ?? 0,
       outputTokens: usage.completion_tokens ?? 0,
-      reasoningTokens: usage.completion_tokens_details?.reasoning_tokens ?? 0,
-      cacheReadTokens: usage.prompt_tokens_details?.cached_tokens ?? 0,
+      reasoningTokens: usage.completion_tokens_details?.reasoning_tokens ?? undefined,
+      cacheReadTokens: usage.prompt_tokens_details?.cached_tokens ?? undefined,
     }),
   })
 }

cloud/app/src/routes/zen/v1/messages.ts

@@ -53,9 +53,9 @@ export function POST(input: APIEvent) {
     normalizeUsage: (usage: Usage) => ({
       inputTokens: usage.input_tokens ?? 0,
       outputTokens: usage.output_tokens ?? 0,
-      cacheReadTokens: usage.cache_read_input_tokens ?? 0,
-      cacheWrite5mTokens: usage.cache_creation?.ephemeral_5m_input_tokens,
-      cacheWrite1hTokens: usage.cache_creation?.ephemeral_1h_input_tokens,
+      cacheReadTokens: usage.cache_read_input_tokens ?? undefined,
+      cacheWrite5mTokens: usage.cache_creation?.ephemeral_5m_input_tokens ?? undefined,
+      cacheWrite1hTokens: usage.cache_creation?.ephemeral_1h_input_tokens ?? undefined,
     }),
   })
 }

cloud/app/src/routes/zen/v1/responses.ts

@@ -1,8 +1,20 @@
 import type { APIEvent } from "@solidjs/start/server"
 import { handler } from "~/util/zen"
 
+type Usage = {
+  input_tokens?: number
+  input_tokens_details?: {
+    cached_tokens?: number
+  }
+  output_tokens?: number
+  output_tokens_details?: {
+    reasoning_tokens?: number
+  }
+  total_tokens?: number
+}
+
 export function POST(input: APIEvent) {
-  let usage: any
+  let usage: Usage
   return handler(input, {
     setAuthHeader: (headers: Headers, apiKey: string) => {
       headers.set("authorization", `Bearer ${apiKey}`)
@@ -15,7 +27,7 @@ export function POST(input: APIEvent) {
 
       let json
       try {
-        json = JSON.parse(data.slice(6))
+        json = JSON.parse(data.slice(6)) as { response?: { usage?: Usage } }
       } catch (e) {
         return
       }
@@ -24,14 +36,14 @@ export function POST(input: APIEvent) {
       usage = json.response.usage
     },
     getStreamUsage: () => usage,
-    normalizeUsage: (usage: any) => {
+    normalizeUsage: (usage: Usage) => {
       const inputTokens = usage.input_tokens ?? 0
       const outputTokens = usage.output_tokens ?? 0
-      const reasoningTokens = usage.output_tokens_details?.reasoning_tokens ?? 0
-      const cacheReadTokens = usage.input_tokens_details?.cached_tokens ?? 0
+      const reasoningTokens = usage.output_tokens_details?.reasoning_tokens ?? undefined
+      const cacheReadTokens = usage.input_tokens_details?.cached_tokens ?? undefined
       return {
-        inputTokens: inputTokens - cacheReadTokens,
-        outputTokens: outputTokens - reasoningTokens,
+        inputTokens: inputTokens - (cacheReadTokens ?? 0),
+        outputTokens: outputTokens - (reasoningTokens ?? 0),
         reasoningTokens,
         cacheReadTokens,
       }

cloud/app/src/util/zen.ts

@@ -10,10 +10,11 @@ import { Resource } from "@opencode/cloud-resource"
 type ModelCost = {
   input: number
   output: number
-  cacheRead: number
-  cacheWrite5m: number
-  cacheWrite1h: number
+  cacheRead?: number
+  cacheWrite5m?: number
+  cacheWrite1h?: number
 }
+
 type Model = {
   id: string
   auth: boolean
@@ -42,7 +43,7 @@ export async function handler(
       inputTokens: number
       outputTokens: number
       reasoningTokens?: number
-      cacheReadTokens: number
+      cacheReadTokens?: number
       cacheWrite5mTokens?: number
       cacheWrite1hTokens?: number
     }
@@ -129,8 +130,6 @@ export async function handler(
         input: 0.00000125,
         output: 0.00001,
         cacheRead: 0.000000125,
-        cacheWrite5m: 0,
-        cacheWrite1h: 0,
       },
       headerMappings: {},
       providers: {
@@ -147,9 +146,6 @@ export async function handler(
       cost: {
         input: 0.00000045,
         output: 0.0000018,
-        cacheRead: 0,
-        cacheWrite5m: 0,
-        cacheWrite1h: 0,
       },
       headerMappings: {},
       providers: {
@@ -173,9 +169,6 @@ export async function handler(
       cost: {
         input: 0.0000006,
         output: 0.0000025,
-        cacheRead: 0,
-        cacheWrite5m: 0,
-        cacheWrite1h: 0,
       },
       headerMappings: {},
       providers: {
@@ -200,8 +193,6 @@ export async function handler(
         input: 0,
         output: 0,
         cacheRead: 0,
-        cacheWrite5m: 0,
-        cacheWrite1h: 0,
       },
       headerMappings: {
         "x-grok-conv-id": "x-opencode-session",
@@ -222,9 +213,6 @@ export async function handler(
       cost: {
         input: 0.00000038,
         output: 0.00000153,
-        cacheRead: 0,
-        cacheWrite5m: 0,
-        cacheWrite1h: 0,
       },
       headerMappings: {},
       providers: {
@@ -438,15 +426,30 @@ export async function handler(
 
       const inputCost = modelCost.input * inputTokens * 100
       const outputCost = modelCost.output * outputTokens * 100
-      const reasoningCost = reasoningTokens ? modelCost.output * reasoningTokens * 100 : undefined
-      const cacheReadCost = modelCost.cacheRead * cacheReadTokens * 100
-      const cacheWrite5mCost = cacheWrite5mTokens ? modelCost.cacheWrite5m * cacheWrite5mTokens * 100 : undefined
-      const cacheWrite1hCost = cacheWrite1hTokens ? modelCost.cacheWrite1h * cacheWrite1hTokens * 100 : undefined
+      const reasoningCost = (() => {
+        if (!reasoningTokens) return undefined
+        return modelCost.output * reasoningTokens * 100
+      })()
+      const cacheReadCost = (() => {
+        if (!cacheReadTokens) return undefined
+        if (!modelCost.cacheRead) return undefined
+        return modelCost.cacheRead * cacheReadTokens * 100
+      })()
+      const cacheWrite5mCost = (() => {
+        if (!cacheWrite5mTokens) return undefined
+        if (!modelCost.cacheWrite5m) return undefined
+        return modelCost.cacheWrite5m * cacheWrite5mTokens * 100
+      })()
+      const cacheWrite1hCost = (() => {
+        if (!cacheWrite1hTokens) return undefined
+        if (!modelCost.cacheWrite1h) return undefined
+        return modelCost.cacheWrite1h * cacheWrite1hTokens * 100
+      })()
       const totalCostInCent =
         inputCost +
         outputCost +
         (reasoningCost ?? 0) +
-        cacheReadCost +
+        (cacheReadCost ?? 0) +
         (cacheWrite5mCost ?? 0) +
         (cacheWrite1hCost ?? 0)
 
@@ -460,7 +463,7 @@ export async function handler(
         "cost.input": Math.round(inputCost),
         "cost.output": Math.round(outputCost),
         "cost.reasoning": reasoningCost ? Math.round(reasoningCost) : undefined,
-        "cost.cache_read": Math.round(cacheReadCost),
+        "cost.cache_read": cacheReadCost ? Math.round(cacheReadCost) : undefined,
         "cost.cache_write_5m": cacheWrite5mCost ? Math.round(cacheWrite5mCost) : undefined,
         "cost.cache_write_1h": cacheWrite1hCost ? Math.round(cacheWrite1hCost) : undefined,
         "cost.total": Math.round(totalCostInCent),
@@ -480,6 +483,8 @@ export async function handler(
           reasoningTokens,
           cacheReadTokens,
           cacheWriteTokens: (cacheWrite5mTokens ?? 0) + (cacheWrite1hTokens ?? 0),
+          cacheWrite5mTokens,
+          cacheWrite1hTokens,
           cost,
         })
         await tx