feat(pricing): add GLM-5.3-Flash pricing, model specs, and catalog registration (#11830)

Adiciona GLM-5.3-Flash ao catálogo com pricing/specs e teste próprio. Validado no worktree combinado (typecheck limpo, teste focado verde). Obrigado!

open-sse/config/glmProvider.ts

@@ -18,6 +18,16 @@ export const GLM_ANTHROPIC_DEFAULT_BASE_URLS = Object.freeze({
 });
 
 export const GLM_SHARED_MODELS = Object.freeze([
+  {
+    id: "glm-5.3-flash",
+    name: "GLM 5.3 Flash",
+    contextLength: 1000000,
+    maxOutputTokens: 131072,
+    toolCalling: true,
+    supportsReasoning: true,
+    supportedThinkingEfforts: ["low", "high", "max"],
+    supportsVision: true,
+  },
   {
     // GLM-5.3 exposes low|high|max reasoning_effort (default max); -high/-low
     // are OmniRoute aliases resolved by GlmExecutor::parseGlmEffortTier.

open-sse/config/providers/registry/zai/index.ts

@@ -19,6 +19,7 @@ export const zaiProvider: RegistryEntry = {
   // unknown IDs. Use the `glm` provider for effort tiers. Vision models are
   // likewise omitted (handled elsewhere).
   models: [
+    { id: "glm-5.3-flash", name: "GLM 5.3 Flash" },
     { id: "glm-5.3", name: "GLM 5.3" },
     { id: "glm-5.2", name: "GLM 5.2" },
     { id: "glm-5.1", name: "GLM 5.1" },

open-sse/mcp-server/__tests__/glmCodingProviderConfig.test.ts

@@ -87,6 +87,7 @@ describe("GLM Coding provider registry surfaces", () => {
     expect(PROVIDER_ID_TO_ALIAS.glm).toBe("glm");
     expect(byProviderId).toEqual(byAlias);
     expect(byProviderId.map((model) => model.id)).toEqual([
+      "glm-5.3-flash",
       "glm-5.3",
       "glm-5.3-high",
       "glm-5.3-low",
@@ -109,6 +110,7 @@ describe("GLM Coding provider registry surfaces", () => {
 
   it("declares exact GLM reasoning-effort tiers across every shared GLM provider", () => {
     const routedTiers = new Map<string, readonly string[]>([
+      ["glm-5.3-flash", ["low", "high", "max"]],
       ["glm-5.3", ["low", "high", "max"]],
       ["glm-5.3-high", ["high"]],
       ["glm-5.3-low", ["low"]],
@@ -181,6 +183,7 @@ describe("GLM Coding provider registry surfaces", () => {
     const get = (id: string) => models.find((m) => m.id === id);
 
     expect(get("glm-5")?.toolCalling).toBe(true);
+    expect(get("glm-5.3-flash")?.toolCalling).toBe(true);
     expect(get("glm-4.7-flash")?.toolCalling).toBe(true);
     expect(get("glm-4.5-air")?.toolCalling).toBe(true);
     expect(get("glm-5.2")?.toolCalling).toBe(true);
@@ -194,6 +197,13 @@ describe("GLM Coding provider registry surfaces", () => {
       reasoning: 4.8,
       cache_creation: 1.0,
     });
+    expect(getPricingForModel("glm", "glm-5.3-flash")).toEqual({
+      input: 0.075,
+      output: 0.25,
+      cached: 0.015,
+      reasoning: 0.25,
+      cache_creation: 0.075,
+    });
     expect(getPricingForModel("glm", "glm-4.7-flash")).toEqual({
       input: 0,
       output: 0,

src/shared/constants/modelSpecs.ts

@@ -67,6 +67,7 @@ const BEDROCK_CLAUDE_ALIASES = (...modelIds: string[]) => [
 // Keep native/bare Z.AI GLM-5.2 context authoritative, but do not blindly apply
 // it to every provider-wrapped alias: hosted providers can and do cap lower.
 const AUTHORITATIVE_CONTEXT_WINDOW_MODEL_IDS = new Set([
+  "glm-5.3-flash",
   "glm-5.3",
   "glm-5.3-high",
   "glm-5.3-low",
@@ -553,6 +554,14 @@ export const MODEL_SPECS: Record<string, ModelSpec> = {
 
   // ── Z.AI GLM-5.3 (1M context mirrored from 5.2 — same base model; 128K max
   // output; effort via reasoning_effort param, tiers are OmniRoute aliases) ──
+  "glm-5.3-flash": {
+    maxOutputTokens: 131072,
+    contextWindow: 1000000,
+    thinkingBudgetCap: 38912,
+    supportsThinking: true,
+    supportsTools: true,
+    supportsVision: true,
+  },
   "glm-5.3": {
     maxOutputTokens: 131072,
     contextWindow: 1000000,

src/shared/constants/pricing/shared-tiers.ts

@@ -111,6 +111,14 @@ export const CLAUDE_SONNET_5_PRICING = {
 };
 
 export const GLM_PRICING = {
+  // GLM-5.3 Flash (2026-08-26, https://models.dev / Z.AI)
+  "glm-5.3-flash": {
+    input: 0.075,
+    output: 0.25,
+    cached: 0.015,
+    reasoning: 0.25,
+    cache_creation: 0.075,
+  },
   // GLM-5.3 (2026-08-14): Z.ai hasn't published 5.3 rates yet — mirrored from
   // GLM-5.2 (same base model; 5.1 and 5.2 also share identical rates).
   // Correct when https://docs.z.ai/guides/overview/pricing lists glm-5.3.

src/shared/constants/visionModels.ts

@@ -38,6 +38,7 @@ export const VISION_MODEL_ID_FRAGMENTS = [
   "glm-4v",
   "glm-4.5v",
   "glm-4.6v",
+  "glm-5.3-flash",
   "gpt-4o",
   "gpt-4.1",
   "gpt-4-turbo",

tests/unit/glm-5.3-catalog-and-effort-tiers.test.ts

@@ -96,6 +96,7 @@ test("catalog suppresses inferred tiers for every GLM registry entry without a p
 
 test("catalog exposes only GLM effort tiers that each provider can route", () => {
   const routedTiers = new Map<string, string[]>([
+    ["glm-5.3-flash", ["low", "high", "max"]],
     ["glm-5.3", ["low", "high", "max"]],
     ["glm-5.3-high", ["high"]],
     ["glm-5.3-low", ["low"]],
@@ -139,9 +140,10 @@ for (const provider of ["glm", "glm-cn", "glmt"]) {
   });
 }
 
-test("zai advertises the GLM-5.3 base model only (DefaultExecutor sends ids verbatim)", () => {
+test("zai advertises the GLM-5.3 base model and GLM-5.3 Flash (DefaultExecutor sends ids verbatim)", () => {
   const ids = modelIds("zai");
   assert.ok(ids.includes("glm-5.3"), `zai should advertise glm-5.3; got ${ids.join(", ")}`);
+  assert.ok(ids.includes("glm-5.3-flash"), `zai should advertise glm-5.3-flash; got ${ids.join(", ")}`);
   for (const alias of ["glm-5.3-high", "glm-5.3-low", "glm-5.3-max"]) {
     assert.ok(
       !ids.includes(alias),
@@ -150,17 +152,18 @@ test("zai advertises the GLM-5.3 base model only (DefaultExecutor sends ids verb
   }
 });
 
-test("modelSpecs carries 1M/128K specs for all GLM-5.3 ids", () => {
-  for (const id of GLM_5_3_IDS) {
+test("modelSpecs carries 1M/128K specs for all GLM-5.3 ids including flash", () => {
+  for (const id of [...GLM_5_3_IDS, "glm-5.3-flash"]) {
     const spec = MODEL_SPECS[id];
     assert.ok(spec, `MODEL_SPECS should include ${id}`);
     assert.equal(spec.contextWindow, 1_000_000);
     assert.equal(spec.maxOutputTokens, 131_072);
     assert.equal(spec.supportsThinking, true);
   }
+  assert.equal(MODEL_SPECS["glm-5.3-flash"]?.supportsVision, true);
 });
 
-test("GLM_PRICING covers the GLM-5.3 ids with GLM-5.2-parity rates", () => {
+test("GLM_PRICING covers the GLM-5.3 ids with GLM-5.2-parity rates and GLM-5.3-Flash rates", () => {
   const reference = GLM_PRICING["glm-5.2"];
   assert.ok(reference, "glm-5.2 pricing reference missing");
   for (const id of GLM_5_3_IDS) {
@@ -168,6 +171,11 @@ test("GLM_PRICING covers the GLM-5.3 ids with GLM-5.2-parity rates", () => {
     assert.ok(pricing, `GLM_PRICING should include ${id}`);
     assert.deepEqual(pricing, reference);
   }
+  const flashPricing = GLM_PRICING["glm-5.3-flash"];
+  assert.ok(flashPricing, "glm-5.3-flash pricing missing");
+  assert.equal(flashPricing.input, 0.075);
+  assert.equal(flashPricing.output, 0.25);
+  assert.equal(flashPricing.cached, 0.015);
 });
 
 test("GlmExecutor resolves glm-5.3-high to reasoning_effort=high on the OpenAI coding transport", () => {