diff --git a/src/agents/context.lookup.test.ts b/src/agents/context.lookup.test.ts
index 87792534270..8f3ba6ee176 100644
--- a/src/agents/context.lookup.test.ts
+++ b/src/agents/context.lookup.test.ts
@@ -358,6 +358,25 @@ describe("lookupContextTokens", () => {
     expect(result).toBe(200_000);
   });
 
+  it("resolveContextTokensForModel treats explicit config as authoritative for read-only misses", async () => {
+    const loadConfig = vi.fn(() => {
+      throw new Error("runtime config should not be loaded");
+    });
+    mockContextModuleDeps(loadConfig);
+    const resolveContextTokensForModel = await importResolveContextTokensForModel();
+
+    const result = resolveContextTokensForModel({
+      cfg: { agents: { defaults: {} } } as never,
+      provider: "openai",
+      model: "unknown-test-model",
+      fallbackContextTokens: 123_000,
+      allowAsyncLoad: false,
+    });
+
+    expect(result).toBe(123_000);
+    expect(loadConfig).not.toHaveBeenCalled();
+  });
+
   it("resolveContextTokensForModel: config direct scan prevents OpenRouter qualified key collision for Google provider", async () => {
     // When provider is explicitly "google" and cfg has a Google contextWindow
     // override, the config direct scan returns it before any cache lookup —
diff --git a/src/agents/context.ts b/src/agents/context.ts
index fe956d151ca..09643fd8588 100644
--- a/src/agents/context.ts
+++ b/src/agents/context.ts
@@ -269,11 +269,14 @@ function ensureContextWindowCacheLoaded(): Promise<void> {
 
 export function lookupContextTokens(
   modelId?: string,
-  options?: { allowAsyncLoad?: boolean },
+  options?: { allowAsyncLoad?: boolean; skipRuntimeConfigLoad?: boolean },
 ): number | undefined {
   if (!modelId) {
     return undefined;
   }
+  if (options?.skipRuntimeConfigLoad) {
+    return lookupCachedContextTokens(modelId);
+  }
   if (options?.allowAsyncLoad === false) {
     // Read-only callers still need synchronous config-backed overrides, but they
     // should not start background model discovery or models.json writes.
@@ -515,7 +518,10 @@ export function resolveContextTokensForModel(params: {
   if (params.provider && ref && !ref.model.includes("/")) {
     const qualifiedResult = lookupContextTokens(
       `${normalizeProviderId(ref.provider)}/${ref.model}`,
-      { allowAsyncLoad: params.allowAsyncLoad },
+      {
+        allowAsyncLoad: params.allowAsyncLoad,
+        skipRuntimeConfigLoad: Boolean(params.cfg),
+      },
     );
     if (qualifiedResult !== undefined) {
       return qualifiedResult;
@@ -526,6 +532,7 @@ export function resolveContextTokensForModel(params: {
   // (e.g. "google/gemini-2.5-pro") this IS the raw discovery cache key.
   const bareResult = lookupContextTokens(params.model, {
     allowAsyncLoad: params.allowAsyncLoad,
+    skipRuntimeConfigLoad: Boolean(params.cfg),
   });
   if (bareResult !== undefined) {
     return bareResult;
@@ -537,7 +544,10 @@ export function resolveContextTokensForModel(params: {
   if (!params.provider && ref && !ref.model.includes("/")) {
     const qualifiedResult = lookupContextTokens(
       `${normalizeProviderId(ref.provider)}/${ref.model}`,
-      { allowAsyncLoad: params.allowAsyncLoad },
+      {
+        allowAsyncLoad: params.allowAsyncLoad,
+        skipRuntimeConfigLoad: Boolean(params.cfg),
+      },
     );
     if (qualifiedResult !== undefined) {
       return qualifiedResult;
diff --git a/src/status/status-message.ts b/src/status/status-message.ts
index 4ba78c3cdb3..a65c56f9a55 100644
--- a/src/status/status-message.ts
+++ b/src/status/status-message.ts
@@ -869,15 +869,16 @@ export function buildStatusMessage(args: StatusArgs): string {
     ? activeAuthMode
     : (selectedAuthMode ?? activeAuthMode);
   const showCost = effectiveCostAuthMode === "api-key" || effectiveCostAuthMode === "mixed";
-  const costConfig = showCost
-    ? resolveModelCostConfig({
-        provider: activeProvider,
-        model: activeModel,
-        config: args.config,
-        allowPluginNormalization: false,
-      })
-    : undefined;
   const hasUsage = typeof inputTokens === "number" || typeof outputTokens === "number";
+  const costConfig =
+    showCost && hasUsage
+      ? resolveModelCostConfig({
+          provider: activeProvider,
+          model: activeModel,
+          config: args.config,
+          allowPluginNormalization: false,
+        })
+      : undefined;
   const cost =
     showCost && hasUsage
       ? estimateUsageCost({