diff --git a/src/agents/context.lookup.test.ts b/src/agents/context.lookup.test.ts
index bcbcd3a9f83..8db58dce89d 100644
--- a/src/agents/context.lookup.test.ts
+++ b/src/agents/context.lookup.test.ts
@@ -366,4 +366,19 @@ describe("lookupContextTokens", () => {
     });
     expect(result).toBe(1_048_576);
   });
+
+  it("resolveContextTokensForModel normalizes explicit provider aliases before config lookup", async () => {
+    mockDiscoveryDeps([]);
+
+    const cfg = createContextOverrideConfig("z.ai", "glm-5", 256_000);
+    const { resolveContextTokensForModel } = await import("./context.js");
+    await flushAsyncWarmup();
+
+    const result = resolveContextTokensForModel({
+      cfg: cfg as never,
+      provider: "z-ai",
+      model: "glm-5",
+    });
+    expect(result).toBe(256_000);
+  });
 });
diff --git a/src/agents/context.ts b/src/agents/context.ts
index 695427936d6..7a589d4d7d9 100644
--- a/src/agents/context.ts
+++ b/src/agents/context.ts
@@ -292,9 +292,11 @@ function resolveProviderModelRef(params: {
   }
   const providerRaw = params.provider?.trim();
   if (providerRaw) {
-    // Keep the exact (lowercased) provider key; callers that need the canonical
-    // alias (e.g. cache key construction) apply normalizeProviderId explicitly.
-    return { provider: providerRaw.toLowerCase(), model: modelRaw };
+    const provider = normalizeProviderId(providerRaw);
+    if (!provider) {
+      return undefined;
+    }
+    return { provider, model: modelRaw };
   }
   const slash = modelRaw.indexOf("/");
   if (slash <= 0) {