fix(chat): apply the Fast capability gate to the native passthrough (#1886)

src/adapters/openai-chat.ts

@@ -14,11 +14,14 @@ import { buildNonOpenAIToolCatalogNudgeForTools, shouldInjectNonOpenAIToolCatalo
 import { openRouterProviderPayload, resolveOpenRouterRouting } from "../providers/openrouter-routing";
 import {
   canForwardForeignServiceTierForChatModel,
+  fastPolicyForModel,
   supportsServiceTierForModel,
 } from "../providers/service-tier";
 import {
   canonicalFastTierMarker,
   createAdapterTierMetadata,
+  decideTier,
+  type ResolvedFastPolicy,
 } from "../providers/fastwire";
 import { openaiChatCompletionsUrl } from "./openai-chat-url";
 import { stripResponsesOnlyEncryptedMarker } from "./responses-tool-schema";
@@ -95,6 +98,8 @@ export function buildOpenAIChatPassthroughRequest(
   rawBody: Record<string, unknown>,
   modelId: string,
   stream: boolean,
+  fastPolicy: ResolvedFastPolicy = fastPolicyForModel(provider, modelId, undefined, "chat"),
+  fastMode?: boolean,
 ): AdapterRequest {
   const { url, headers, hasCredential } = openAIChatTransport(provider);
 
@@ -122,7 +127,16 @@ export function buildOpenAIChatPassthroughRequest(
   // `<listed>:<tag>` siblings the operator never opted out, silently returning prose.
   if (provider.noStructuredOutputModels?.includes(modelId)) delete body.response_format;
 
-  if (provider.chatServiceTier && rawBody.service_tier !== undefined) {
+  // Run the same complete Fast policy as the translated Chat path, including explicit
+  // fastMode and foreign-tier handling. On inherited canonical Fast, the passthrough still
+  // retains the caller's exact spelling; forced Fast uses the policy-owned wire value.
+  const callerTier = typeof rawBody.service_tier === "string" ? rawBody.service_tier : undefined;
+  const tierDecision = decideTier(fastPolicy, fastMode, callerTier);
+  if (tierDecision.kind === "set") {
+    body.service_tier = fastMode === undefined && canonicalFastTierMarker(callerTier) !== undefined
+      ? callerTier
+      : tierDecision.value;
+  } else if (tierDecision.kind === "forward-caller" && rawBody.service_tier !== undefined) {
     body.service_tier = rawBody.service_tier;
   }
   if (provider.promptCacheKey && rawBody.prompt_cache_key !== undefined) {
@@ -1284,6 +1298,23 @@ function thinkingBudgetForEffort(parsed: OcxParsedRequest, reasoningEffort: stri
   return fraction === undefined ? undefined : Math.max(1, Math.floor(maxBudget * fraction));
 }
 
+function canSerializeOpenAIChatServiceTier(
+  provider: OcxProviderConfig,
+  modelId: string,
+  serviceTier: unknown,
+  tierDecision?: OcxParsedRequest["options"]["tierDecision"],
+): boolean {
+  if (serviceTier === undefined) return false;
+  if (tierDecision !== undefined) {
+    return tierDecision.kind === "set" || tierDecision.kind === "forward-caller";
+  }
+  const callerTier = typeof serviceTier === "string" ? serviceTier : undefined;
+  const callerCanonicalFast = canonicalFastTierMarker(callerTier) !== undefined;
+  const capability = supportsServiceTierForModel(provider, modelId);
+  const callerTierForwardAllowed = canForwardForeignServiceTierForChatModel(provider, modelId);
+  return callerTierForwardAllowed || (callerCanonicalFast && capability === true);
+}
+
 export function createOpenAIChatAdapter(provider: OcxProviderConfig): ProviderAdapter {
   return {
     name: "openai-chat",
@@ -1306,13 +1337,12 @@ export function createOpenAIChatAdapter(provider: OcxProviderConfig): ProviderAd
       // unclassified Chat routes remain behind the caller-forwarding opt-in.
       const serviceTier = parsed.options.serviceTier;
       const tierDecision = parsed.options.tierDecision;
-      const callerCanonicalFast = canonicalFastTierMarker(serviceTier) !== undefined;
-      const callerTierForwardAllowed = canForwardForeignServiceTierForChatModel(provider, parsed.modelId);
-      const canonicalFastCapability = callerCanonicalFast
-        && supportsServiceTierForModel(provider, parsed.modelId) === true;
-      const canSerializeServiceTier = tierDecision?.kind === "set"
-        || tierDecision?.kind === "forward-caller"
-        || (tierDecision === undefined && (callerTierForwardAllowed || canonicalFastCapability));
+      const canSerializeServiceTier = canSerializeOpenAIChatServiceTier(
+        provider,
+        parsed.modelId,
+        serviceTier,
+        tierDecision,
+      );
       if (canSerializeServiceTier && serviceTier !== undefined) {
         body.service_tier = serviceTier;
       }

src/server/chat-native.ts

@@ -27,6 +27,7 @@ import {
   rateLimitRetryPolicyFor,
   rotateProviderTransportOn429,
 } from "../providers/key-failover";
+import { fastPolicyForModel } from "../providers/service-tier";
 import type { RouteResult } from "../router";
 import type { OcxConfig, OcxProviderConfig } from "../types";
 import { fetchWithHeaderTimeout, providerFetch, safeHostLabel } from "./responses/fetch-helpers";
@@ -154,8 +155,16 @@ export async function handleNativeChatCompletions(options: HandleNativeChatOptio
     translatorBudget.chargeRetained(bytes, { kind: "request_copies" });
     retainedRequestBytes = bytes;
   };
+  const buildActiveRequest = () => buildOpenAIChatPassthroughRequest(
+    activeProvider,
+    options.chatBody,
+    route.modelId,
+    requestedStream,
+    fastPolicyForModel(activeProvider, route.modelId, route.providerName, "chat"),
+    config.fastMode,
+  );
   try {
-    activeRequest = buildOpenAIChatPassthroughRequest(activeProvider, options.chatBody, route.modelId, requestedStream);
+    activeRequest = buildActiveRequest();
     retainRequest(activeRequest);
   } catch (error) {
     releaseRetainedRequest();
@@ -222,7 +231,7 @@ export async function handleNativeChatCompletions(options: HandleNativeChatOptio
       activeProvider = rotated;
       activeAdapter = createOpenAIChatAdapter(activeProvider);
       releaseRetainedRequest();
-      activeRequest = buildOpenAIChatPassthroughRequest(activeProvider, options.chatBody, route.modelId, requestedStream);
+      activeRequest = buildActiveRequest();
       retainRequest(activeRequest);
       response = await send(activeRequest, "key-429");
     }

tests/fastwire-characterization-wire.test.ts

@@ -1,6 +1,7 @@
 import { afterEach, describe, expect, spyOn, test } from "bun:test";
 import { buildOpenAIChatPassthroughRequest } from "../src/adapters/openai-chat";
 import { chatCompletionsToResponsesBody } from "../src/chat/inbound";
+import { fastPolicyForModel } from "../src/providers/service-tier";
 import * as adapterResolveModule from "../src/server/adapter-resolve";
 import type { RequestLogContext } from "../src/server/request-log";
 import { handleResponses } from "../src/server/responses/core";
@@ -364,27 +365,30 @@ describe("FastWire characterization: rawBody observation point", () => {
 });
 
 describe("FastWire characterization: known bugs", () => {
-  test("characterization (known bug): native chat passthrough ignores exact-model false", () => {
+  test("characterization: native chat passthrough honors exact-model false", () => {
+    // FastWire #1886 native-chat policy fix: exact-model false now strips the caller tier.
+    const provider: OcxProviderConfig = {
+      adapter: "openai-chat",
+      baseUrl: "https://native-chat.example.test/v1",
+      authMode: "key",
+      apiKey: "sk-test",
+      supportsServiceTier: true,
+      chatServiceTier: true,
+      modelSupportsServiceTier: { model: false },
+    };
     const request = buildOpenAIChatPassthroughRequest(
-      {
-        adapter: "openai-chat",
-        baseUrl: "https://native-chat.example.test/v1",
-        authMode: "key",
-        apiKey: "sk-test",
-        supportsServiceTier: true,
-        chatServiceTier: true,
-        modelSupportsServiceTier: { model: false },
-      },
+      provider,
       {
         model: "model",
         messages: [{ role: "user", content: "ping" }],
         service_tier: "flex",
       },
       "model",
       false,
+      fastPolicyForModel(provider, "model", "native-chat", "chat"),
     );
     const body = JSON.parse(request.body) as Record<string, unknown>;
-    expect(body.service_tier).toBe("flex");
+    expect(body).not.toHaveProperty("service_tier");
   });
 
   test("characterization: chat-to-responses conversion preserves service_tier", () => {