diff --git a/packages/cloud/shared/src/lib/services/ai-pricing/lookup-fallback-pricing.test.ts b/packages/cloud/shared/src/lib/services/ai-pricing/lookup-fallback-pricing.test.ts
index ae25e1050643b..31a3bcf2a11de 100644
--- a/packages/cloud/shared/src/lib/services/ai-pricing/lookup-fallback-pricing.test.ts
+++ b/packages/cloud/shared/src/lib/services/ai-pricing/lookup-fallback-pricing.test.ts
@@ -12,9 +12,8 @@
  *   3. A provider with no catalogued entries falls back to the env-configured
  *      defaults AI_PRICING_FALLBACK_INPUT_USD_PER_M /
  *      AI_PRICING_FALLBACK_OUTPUT_USD_PER_M.
- *   4. With no catalog and no env default, the request stays servable at a
- *      non-zero hardcoded frontier-max last resort — never $0 (#11635; also
- *      covered in lookup-missing-pricing.test.ts).
+ *   4. With no catalog and no env default, a non-zero token side rejects
+ *      instead of guessing a price. A `..._USD_PER_M=0` env is treated as unset.
  *   5. Reserve (estimated tokens) and settle (actual tokens) resolve the same
  *      fallback rate, so billing stays consistent across the request.
  */
@@ -179,33 +178,56 @@ test("env-configured default applies when the provider has no catalogued entries
   expect(result.totalCost).toBe(12.5 * 1.2);
 });
 
-test("invalid env default is ignored (falls through to the non-zero last resort, #11635)", async () => {
+test("invalid env default is ignored, then missing pricing fails closed (#11635)", async () => {
   process.env.AI_PRICING_FALLBACK_INPUT_USD_PER_M = "not-a-number";
   process.env.AI_PRICING_FALLBACK_OUTPUT_USD_PER_M = "-4";
 
-  const result = await calculateTextCostFromCatalog({
-    model: "mystery-model-1",
-    provider: "someprovider",
-    inputTokens: 1_000,
-    outputTokens: 1_000,
-  });
+  await expect(
+    calculateTextCostFromCatalog({
+      model: "mystery-model-1",
+      provider: "someprovider",
+      inputTokens: 1_000,
+      outputTokens: 1_000,
+    }),
+  ).rejects.toThrow("refusing to bill unknown-priced inference");
+});
 
-  // Invalid env is still ignored, but the last resort is now the hardcoded
-  // frontier-max rate, never $0 (#11635).
-  expect(result.totalCost).toBeGreaterThan(0);
-  expect(result.totalCost).toBeLessThan(0.1);
+test("AI_PRICING_FALLBACK_*=0 is treated as unset, not as a $0 price (#11635)", async () => {
+  process.env.AI_PRICING_FALLBACK_INPUT_USD_PER_M = "0";
+  process.env.AI_PRICING_FALLBACK_OUTPUT_USD_PER_M = "0";
+
+  await expect(
+    calculateTextCostFromCatalog({
+      model: "mystery-model-1",
+      provider: "someprovider",
+      inputTokens: 1_000,
+      outputTokens: 1_000,
+    }),
+  ).rejects.toThrow("refusing to bill unknown-priced inference");
+});
+
+test("no catalog and no env default rejects instead of selling unknown-priced inference (#11635)", async () => {
+  await expect(
+    calculateTextCostFromCatalog({
+      model: "mystery-model-1",
+      provider: "someprovider",
+      inputTokens: 1_000,
+      outputTokens: 1_000,
+    }),
+  ).rejects.toThrow("refusing to bill unknown-priced inference");
 });
 
-test("no catalog and no env default keeps the request servable at a non-zero last resort (#11635)", async () => {
+test("a side with zero tokens does not require pricing (#11635)", async () => {
+  process.env.AI_PRICING_FALLBACK_INPUT_USD_PER_M = "2.5";
+
   const result = await calculateTextCostFromCatalog({
     model: "mystery-model-1",
     provider: "someprovider",
     inputTokens: 1_000,
-    outputTokens: 1_000,
+    outputTokens: 0,
   });
 
-  expect(result.inputCost).toBeGreaterThan(0);
-  expect(result.outputCost).toBeGreaterThan(0);
-  expect(result.totalCost).toBeGreaterThan(0);
-  expect(result.totalCost).toBeLessThan(0.1);
+  expect(result.baseOutputCost).toBe(0);
+  expect(result.baseInputCost).toBeCloseTo(0.0025, 9);
+  expect(result.totalCost).toBeCloseTo(0.003, 9);
 });
diff --git a/packages/cloud/shared/src/lib/services/ai-pricing/lookup-missing-pricing.test.ts b/packages/cloud/shared/src/lib/services/ai-pricing/lookup-missing-pricing.test.ts
index f7f5d90bb197a..ae4403c8ad9bf 100644
--- a/packages/cloud/shared/src/lib/services/ai-pricing/lookup-missing-pricing.test.ts
+++ b/packages/cloud/shared/src/lib/services/ai-pricing/lookup-missing-pricing.test.ts
@@ -1,5 +1,5 @@
 /**
- * Regression: a missing price must (1) NOT throw a 500 and (2) NEVER bill $0.
+ * Regression: a missing price must never bill $0 or a guessed hardcoded floor.
  *
  * `calculateTextCostFromCatalog` once left the input-price lookup unguarded, so
  * a catalog miss threw `Pricing unavailable for language:input <model>` → a 500
@@ -9,14 +9,15 @@
  * inference / uncollected revenue (#11635).
  *
  * Both seams (persisted repo + live gateway) are mocked empty so EVERY lookup
- * misses — the worst case — and no env fallback is set, so the last-resort tier
- * is exercised. Post-#11635 it bills a conservative non-zero frontier-max rate
- * (`lastResortTokenUnitPrice`), keyed by product family, and still never throws.
- * (Provider-max / env-default tiers are covered in lookup-fallback-pricing.test.ts.)
+ * misses — the worst case — and no env fallback is set. Post-#11635, a non-zero
+ * token side rejects so we do not sell inference we do not know how to price.
+ * Provider-max and env-default fallback tiers are covered in
+ * lookup-fallback-pricing.test.ts.
  */
 import { beforeEach, expect, mock, test } from "bun:test";
 
 const warnSpy = mock(() => {});
+const errorSpy = mock(() => {});
 
 mock.module("../../../db/repositories/ai-pricing", () => ({
   aiPricingRepository: {
@@ -27,6 +28,7 @@ mock.module("../../../db/repositories/ai-pricing", () => ({
 mock.module("../../utils/logger", () => ({
   logger: {
     warn: warnSpy,
+    error: errorSpy,
   },
 }));
 mock.module("./providers/gateway", () => ({
@@ -37,56 +39,53 @@ const { calculateTextCostFromCatalog } = await import("./lookup");
 
 beforeEach(() => {
   warnSpy.mockClear();
+  errorSpy.mockClear();
 });
 
-test("missing language pricing bills a non-zero last-resort rate, never $0 (#11635)", async () => {
-  const result = await calculateTextCostFromCatalog({
-    model: "totally-uncatalogued-model",
-    provider: "someprovider",
-    inputTokens: 1000,
-    outputTokens: 500,
-  });
+test("missing language pricing rejects instead of billing an unknown price (#11635)", async () => {
+  await expect(
+    calculateTextCostFromCatalog({
+      model: "totally-uncatalogued-model",
+      provider: "someprovider",
+      inputTokens: 1000,
+      outputTokens: 500,
+    }),
+  ).rejects.toThrow("refusing to bill unknown-priced inference");
 
-  // Never throws (the original 500-degradation) AND never $0 (the #11635 fix):
-  // 1000in@$5/M + 500out@$25/M ≈ $0.0175 base + markup — a small, sane amount.
-  expect(result.inputCost).toBeGreaterThan(0);
-  expect(result.outputCost).toBeGreaterThan(0);
-  expect(result.totalCost).toBeGreaterThan(0);
-  expect(result.totalCost).toBeLessThan(0.1); // sane: not an absurd rate
-  expect(warnSpy.mock.calls).toContainEqual([
-    "ai-pricing: input pricing unavailable; billing at fallback rate",
+  expect(errorSpy.mock.calls).toContainEqual([
+    "ai-pricing: missing token price with no fallback; refusing request",
     {
       canonicalModel: "someprovider/totally-uncatalogued-model",
       provider: "someprovider",
       billingSource: undefined,
-      fallbackSource: "last_resort",
-      fallbackUnitPrice: 0.000005,
+      productFamily: "language",
+      chargeType: "input",
+      tokens: 1000,
     },
   ]);
-  expect(warnSpy.mock.calls).toContainEqual([
-    "ai-pricing: output pricing unavailable; billing at fallback rate",
+  expect(warnSpy.mock.calls).toHaveLength(0);
+});
+
+test("missing input-only embedding pricing rejects instead of billing an unknown price (#11635)", async () => {
+  await expect(
+    calculateTextCostFromCatalog({
+      model: "uncatalogued-embedding-model",
+      provider: "someprovider",
+      inputTokens: 800,
+      outputTokens: 0,
+    }),
+  ).rejects.toThrow("refusing to bill unknown-priced inference");
+
+  expect(errorSpy.mock.calls).toContainEqual([
+    "ai-pricing: missing token price with no fallback; refusing request",
     {
-      canonicalModel: "someprovider/totally-uncatalogued-model",
+      canonicalModel: "someprovider/uncatalogued-embedding-model",
       provider: "someprovider",
       billingSource: undefined,
-      fallbackSource: "last_resort",
-      fallbackUnitPrice: 0.000025,
+      productFamily: "embedding",
+      chargeType: "input",
+      tokens: 800,
     },
   ]);
-});
-
-test("missing embedding pricing bills the cheaper embedding last-resort rate, non-zero (#11635)", async () => {
-  const result = await calculateTextCostFromCatalog({
-    model: "uncatalogued-embedding-model",
-    provider: "someprovider",
-    inputTokens: 800,
-    outputTokens: 0,
-  });
-
-  // Non-zero (never free) but keyed to the embedding family ($0.2/M), so 800
-  // tokens is a tiny amount — proving the family keying picked the cheap rate,
-  // not the $5/M language default.
-  expect(result.inputCost).toBeGreaterThan(0);
-  expect(result.totalCost).toBeGreaterThan(0);
-  expect(result.totalCost).toBeLessThan(0.001);
+  expect(warnSpy.mock.calls).toHaveLength(0);
 });
diff --git a/packages/cloud/shared/src/lib/services/ai-pricing/lookup.ts b/packages/cloud/shared/src/lib/services/ai-pricing/lookup.ts
index 1d3290ce58c63..2ddb4840e20c2 100644
--- a/packages/cloud/shared/src/lib/services/ai-pricing/lookup.ts
+++ b/packages/cloud/shared/src/lib/services/ai-pricing/lookup.ts
@@ -169,7 +169,10 @@ function envFallbackTokenUnitPrice(chargeType: "input" | "output"): number | nul
     return null;
   }
   const usdPerMillion = Number(raw);
-  if (!Number.isFinite(usdPerMillion) || usdPerMillion < 0) {
+  // #11635: reject 0 too (not just negative/non-finite) — a `..._USD_PER_M=0`
+  // env value would otherwise masquerade as a configured floor while still
+  // billing $0. Treat it as unset so the missing-price path fails closed.
+  if (!Number.isFinite(usdPerMillion) || usdPerMillion <= 0) {
     logger.warn("ai-pricing: ignoring invalid fallback-rate env value", {
       envName,
       value: raw,
@@ -179,33 +182,9 @@ function envFallbackTokenUnitPrice(chargeType: "input" | "output"): number | nul
   return usdPerMillion / 1_000_000;
 }
 
-/**
- * Fail-closed last-resort per-token rate (USD/token) for a servable but
- * uncatalogued model with no provider-max catalog entry AND no env fallback
- * (#11635). The request still serves — it just can never bill $0. Deliberately
- * conservative (frontier-max) so an unpriced model over-bills rather than gives
- * away free inference; a real catalog entry (cheaper) supersedes this on the
- * next lookup, and the loud log points the catalog gap.
- */
-const LAST_RESORT_USD_PER_MILLION: Partial<
-  Record<PricingProductFamily, { input: number; output: number }>
-> = {
-  language: { input: 5, output: 25 },
-  embedding: { input: 0.2, output: 0.2 },
-};
-const DEFAULT_LAST_RESORT_USD_PER_MILLION = { input: 5, output: 25 };
-
-function lastResortTokenUnitPrice(
-  productFamily: PricingProductFamily,
-  chargeType: "input" | "output",
-): number {
-  const family = LAST_RESORT_USD_PER_MILLION[productFamily] ?? DEFAULT_LAST_RESORT_USD_PER_MILLION;
-  return family[chargeType] / 1_000_000;
-}
-
 type FallbackTokenRate = {
   unitPrice: number;
-  source: "provider_max_catalog" | "env_default" | "last_resort";
+  source: "provider_max_catalog" | "env_default";
   referenceModel?: string;
 };
 
@@ -219,10 +198,8 @@ type FallbackTokenRate = {
  * catalogued token rate for the same product family/charge type (an upper
  * bound over any plausible real price from that provider), or an
  * env-configured default (AI_PRICING_FALLBACK_{INPUT,OUTPUT}_USD_PER_M) when
- * the provider has no catalogued entries at all, or a hardcoded last-resort
- * rate when neither source exists. Both reserve (pre-flight estimate) and
- * settle (actual usage) resolve through this same path, so both sides of a
- * request bill at the same rate.
+ * the provider has no catalogued entries at all. If neither source exists, the
+ * caller must fail closed rather than inventing a price.
  */
 async function resolveFallbackTokenRate(params: {
   billingSource?: PricingBillingSource;
@@ -294,10 +271,7 @@ async function resolveFallbackTokenRate(params: {
     return { unitPrice: envUnitPrice, source: "env_default" };
   }
 
-  return {
-    unitPrice: lastResortTokenUnitPrice(params.productFamily, params.chargeType),
-    source: "last_resort",
-  };
+  return null;
 }
 
 function computeCostFromEntry(entry: PreparedPricingEntry, quantity: number): FlatOperationCost {
@@ -373,10 +347,9 @@ export async function calculateTextCostFromCatalog(params: {
   // the catalog (notably embedding models, which are input-only and run every
   // turn). A servable request must never fail purely on a missing price — but
   // it must not be under-billed at $0 either. On a miss, bill the missing side
-  // at a conservative fallback rate (provider max → env default → a non-zero
-  // hardcoded frontier-max last resort, NEVER $0; see resolveFallbackTokenRate
-  // + lastResortTokenUnitPrice, #11635) and log loudly so the catalog gap gets
-  // priced.
+  // only when a real fallback exists (provider max → env default). If neither
+  // source exists, fail closed because we should not sell inference we do not
+  // know how to price (#11635).
   const inputEntry = await resolvePreparedPricingEntry({
     billingSource: params.billingSource,
     provider: params.provider,
@@ -409,13 +382,25 @@ export async function calculateTextCostFromCatalog(params: {
       productFamily,
       chargeType,
     });
+    if (!fallback) {
+      const message = `Pricing unavailable for ${productFamily}:${chargeType} ${canonicalModel}; refusing to bill unknown-priced inference`;
+      logger.error("ai-pricing: missing token price with no fallback; refusing request", {
+        canonicalModel,
+        provider: params.provider,
+        billingSource: params.billingSource,
+        productFamily,
+        chargeType,
+        tokens,
+      });
+      throw new Error(message);
+    }
     logger.warn(`ai-pricing: ${chargeType} pricing unavailable; billing at fallback rate`, {
       canonicalModel,
       provider: params.provider,
       billingSource: params.billingSource,
-      fallbackSource: fallback?.source ?? "none",
-      fallbackUnitPrice: fallback?.unitPrice ?? 0,
-      ...(fallback?.referenceModel ? { fallbackReferenceModel: fallback.referenceModel } : {}),
+      fallbackSource: fallback.source,
+      fallbackUnitPrice: fallback.unitPrice,
+      ...(fallback.referenceModel ? { fallbackReferenceModel: fallback.referenceModel } : {}),
     });
     return fallback;
   };
@@ -427,10 +412,10 @@ export async function calculateTextCostFromCatalog(params: {
 
   const inputUnitPrice = inputEntry
     ? asDecimal(inputEntry.unitPrice)
-    : asDecimal(inputFallback?.unitPrice ?? lastResortTokenUnitPrice(productFamily, "input"));
+    : asDecimal(inputFallback?.unitPrice ?? 0);
   const outputUnitPrice = outputEntry
     ? asDecimal(outputEntry.unitPrice)
-    : asDecimal(outputFallback?.unitPrice ?? lastResortTokenUnitPrice(productFamily, "output"));
+    : asDecimal(outputFallback?.unitPrice ?? 0);
 
   const baseInputCost = inputUnitPrice.mul(params.inputTokens);
   const baseOutputCost = outputUnitPrice.mul(params.outputTokens);