diff --git a/CHANGELOG.md b/CHANGELOG.md
index 52fd17a945..f2ff073a2f 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -74,6 +74,17 @@
 
 ### ✅ Tests
 
+- **Combo strategy fallback coverage** (`tests/unit/combo-strategy-fallbacks.test.ts`, 11 tests): fill-first / p2c / random / cost-optimized / strict-random fallback paths (previously happy-path only), price-tie stability, stale strict-random deck degradation, unknown-strategy normalization to priority, and circuit-breaker HALF_OPEN recovery inside the combo loop + `preScreenTargets` (lazy-recovery contract).
+- **`#1731` fast-skip suite restored** (`tests/integration/combo-provider-exhaustion.test.ts`): the five skipped tests were rewritten against the current routing policy (quota-exhausted 429 marks the provider for the request; transient 429 retries other connections; connection errors skip per-connection; nothing persists across requests) and re-enabled — 8/8 green.
+- **Proxy context passthrough** (`tests/integration/proxy-context-passthrough.test.ts`): combo targets each execute under their own connection's proxy; `count_tokens` runs inside the connection's proxy context.
+### 🐛 Fixed
+
+- chore(quality-gate): reconcile check-file-size baseline — freeze 27 inherited/this-round grown files at current LOC + register providerLimits.ts (greens the file-size gate; shrink tracked via #3501)
+- fix(gemini): standard Gemini provider no longer sends tool calls without a thought_signature (falls back to context mode when the signature is unavailable), fixing HTTP 400 on multi-turn thinking-model tool calls (#3688)
+- fix(antigravity): preserve gemini-3.1-pro High/Low budget tiers (upstream accepts the suffixed ids; stop collapsing to bare gemini-3.1-pro) (#3696)
+- fix: streaming combos now fail over to the next target when an upstream returns an empty/content-filtered response instead of surfacing a blank reply (#3685)
+- fix(qwen-web): migrate the Qwen Web provider to the v2 chat API. The legacy `/api/chat/completions` endpoint was retired upstream and returned `504` HTML from Alibaba's gateway for every request regardless of credentials, so no token refresh could help (#3288, discussion #2768). The executor now uses the two-step v2 flow (`/api/v2/chats/new` → `/api/v2/chat/completions?chat_id=`), replays the full browser cookie jar (cna + ssxmod_itna/itna2 + token) required by Alibaba's WAF instead of only a Bearer token, parses the phase-based SSE (think→reasoning, answer→content), and maps the WAF/expired HTML page to a clear actionable auth error instead of surfacing raw HTML. The credential requirement is now a full Cookie header, browser-login capture grabs the WAF cookies, and the model catalog is refreshed to the current upstream ids (`qwen3.7-max`, `qwen3.7-plus`, `qwen3.6-plus`; legacy ids kept as aliases). 17 unit tests.
+- fix(cache): semantic cache responses are now scoped to the requesting API key — two callers with different API keys that send the same prompt and model will no longer receive each other's cached responses (#3740). `generateSignature` includes the `api_key_id` dimension in the SHA-256 hash; unauthenticated requests (no API key) remain isolated from keyed requests. Existing cache entries (generated without the key dimension) are cleared by migration `098`.
 - **Combo strategy fallback coverage** (`tests/unit/combo-strategy-fallbacks.test.ts`, 11 tests — thanks @diegosouzapw): fill-first / p2c / random / cost-optimized / strict-random fallback paths (previously happy-path only), price-tie stability, stale strict-random deck degradation, unknown-strategy normalization to priority, and circuit-breaker HALF_OPEN recovery inside the combo loop + `preScreenTargets` (lazy-recovery contract).
 - **`#1731` fast-skip suite restored** (`tests/integration/combo-provider-exhaustion.test.ts` — thanks @diegosouzapw): 5 previously-skipped integration tests rewritten against the current routing policy — 8/8 green.
 - **Proxy context passthrough** (`tests/integration/proxy-context-passthrough.test.ts` — thanks @diegosouzapw): combo targets each execute under their own connection's proxy; `count_tokens` runs inside the connection's proxy context.
diff --git a/open-sse/handlers/chatCore.ts b/open-sse/handlers/chatCore.ts
index c5f5188619..c2bf1943f3 100644
--- a/open-sse/handlers/chatCore.ts
+++ b/open-sse/handlers/chatCore.ts
@@ -2292,6 +2292,7 @@ export async function handleChatCore({
     startTime,
     log,
     persistAttemptLogs,
+    apiKeyId: apiKeyInfo?.id ?? undefined,
   });
   if (cacheHit) {
     return cacheHit;
@@ -5186,7 +5187,8 @@ export async function handleChatCore({
         model,
         body.messages ?? body.input,
         body.temperature,
-        body.top_p
+        body.top_p,
+        apiKeyInfo?.id ?? undefined
       );
       const tokensSaved = usage?.prompt_tokens + usage?.completion_tokens || 0;
       setCachedResponse(signature, model, translatedResponse, tokensSaved);
@@ -5580,7 +5582,8 @@ export async function handleChatCore({
           model,
           body.messages ?? body.input,
           body.temperature,
-          body.top_p
+          body.top_p,
+          apiKeyInfo?.id ?? undefined
         );
         const u = streamUsage as Record<string, unknown> | null;
         const tokensSaved =
diff --git a/open-sse/handlers/chatCore/semanticCache.ts b/open-sse/handlers/chatCore/semanticCache.ts
index 9ce133c1a2..6e99989e02 100644
--- a/open-sse/handlers/chatCore/semanticCache.ts
+++ b/open-sse/handlers/chatCore/semanticCache.ts
@@ -23,6 +23,7 @@ export async function checkSemanticCache({
   startTime,
   log,
   persistAttemptLogs,
+  apiKeyId,
 }: {
   semanticCacheEnabled: boolean;
   body: Record<string, unknown>;
@@ -36,13 +37,15 @@ export async function checkSemanticCache({
   startTime: number;
   log: unknown;
   persistAttemptLogs: (args: unknown) => void;
+  apiKeyId?: string | null;
 }) {
   if (semanticCacheEnabled && isCacheableForRead(body, clientRawRequest?.headers)) {
     const signature = generateSignature(
       model,
       body.messages ?? body.input,
       body.temperature,
-      body.top_p
+      body.top_p,
+      apiKeyId ?? undefined
     );
     const cached = getCachedResponse(signature);
     if (cached) {
diff --git a/src/lib/db/migrations/098_clear_semantic_cache_for_key_isolation.sql b/src/lib/db/migrations/098_clear_semantic_cache_for_key_isolation.sql
new file mode 100644
index 0000000000..92e0570305
--- /dev/null
+++ b/src/lib/db/migrations/098_clear_semantic_cache_for_key_isolation.sql
@@ -0,0 +1,4 @@
+-- Migration 098: Clear semantic_cache after key-isolation fix (#3740)
+-- Old signatures were computed without the API key ID dimension, so existing
+-- entries would be shared across users. Truncating forces new scoped entries.
+DELETE FROM semantic_cache;
diff --git a/src/lib/semanticCache.ts b/src/lib/semanticCache.ts
index 1fbc6a6548..b8ddea8a07 100644
--- a/src/lib/semanticCache.ts
+++ b/src/lib/semanticCache.ts
@@ -113,14 +113,16 @@ function getMemoryCache() {
  * @param {Array} messages - Normalized messages array
  * @param {number} temperature
  * @param {number} topP
+ * @param {string} [apiKeyId] - API key ID for per-key isolation (prevents cross-user cache hits)
  * @returns {string} hex signature
  */
-export function generateSignature(model, conversation, temperature = 0, topP = 1) {
+export function generateSignature(model, conversation, temperature = 0, topP = 1, apiKeyId?: string) {
   const payload = JSON.stringify({
     model,
     messages: normalizeConversation(conversation),
     temperature,
     top_p: topP,
+    ...(apiKeyId ? { api_key_id: apiKeyId } : {}),
   });
   return crypto.createHash("sha256").update(payload).digest("hex");
 }
diff --git a/tests/unit/semantic-cache.test.ts b/tests/unit/semantic-cache.test.ts
index 69185613a0..2205594b17 100644
--- a/tests/unit/semantic-cache.test.ts
+++ b/tests/unit/semantic-cache.test.ts
@@ -71,6 +71,29 @@ describe("Semantic Cache", () => {
       const sig2 = generateSignature("gpt-5", input2, 0, 1);
       assert.equal(sig1, sig2);
     });
+
+    // #3740: cross-user cache isolation — different API keys must not share cached responses
+    it("generates different signatures for different API key IDs (#3740)", () => {
+      const messages = [{ role: "user", content: "what is 2+2?" }];
+      const sigKeyA = generateSignature("gpt-4o", messages, 0, 1, "key-id-alice");
+      const sigKeyB = generateSignature("gpt-4o", messages, 0, 1, "key-id-bob");
+      assert.notEqual(sigKeyA, sigKeyB, "different API keys must produce different cache signatures");
+    });
+
+    it("generates consistent signatures for same API key ID (#3740)", () => {
+      const messages = [{ role: "user", content: "what is 2+2?" }];
+      const sig1 = generateSignature("gpt-4o", messages, 0, 1, "key-id-alice");
+      const sig2 = generateSignature("gpt-4o", messages, 0, 1, "key-id-alice");
+      assert.equal(sig1, sig2);
+    });
+
+    it("matches keyless signature when apiKeyId is undefined (#3740)", () => {
+      const messages = [{ role: "user", content: "hello" }];
+      // Unauthenticated requests (apiKeyId=undefined) must not collide with keyed requests
+      const sigKeyed = generateSignature("gpt-4o", messages, 0, 1, "some-key-id");
+      const sigKeyless = generateSignature("gpt-4o", messages, 0, 1, undefined);
+      assert.notEqual(sigKeyed, sigKeyless);
+    });
   });
 
   describe("isCacheableForRead", () => {