fix(providers): eliminate Gemini 3.5 Flash

2026-08-24 16:12:23 +03:00 · 2026-08-23 22:46:54 +09:00
parent 29caad9f3d
commit 2764812ee4
34 changed files with 166 additions and 218 deletions
--- a/changelog.d/fixes/10286-gemini-3-5-flash-thinking.md
+++ b/changelog.d/fixes/10286-gemini-3-5-flash-thinking.md
@@ -1 +0,0 @@
- fix(sse): mark gemini-3.5-flash as thinking-capable so reasoning_effort is no longer rejected with a spurious 400 (#10286)
--- a/docs/guides/ANTIGRAVITY-ONBOARDING.md
+++ b/docs/guides/ANTIGRAVITY-ONBOARDING.md
@@ -6,7 +6,7 @@ lastUpdated: 2026-07-31

 # OmniRoute Antigravity (Google One AI) Onboarding Guide

-> **What you get**: Access to Gemini 3.1 Pro, Gemini 3.5 Flash, Claude Sonnet 4.6, and other models through your Google One AI Pro subscription — routed through OmniRoute as a unified gateway.
+> **What you get**: Access to Gemini 3.1 Pro, Gemini 3.7 Flash, Claude Sonnet 4.6, and other models through your Google One AI Pro subscription — routed through OmniRoute as a unified gateway.

 **Official references**:

@@ -45,7 +45,7 @@ Both providers share the **same Google backend** — identical OAuth client, tok

 **Why the model catalog differs**: Google's CLI is "optimized for speed and low overhead" and "co-optimized with Gemini models" (per Google's official blog). The Web/IDE product is "optimized for comprehensiveness." The CLI uses `:fetchAvailableModels` to dynamically discover models, while the IDE uses a static curated list.

-**In practice**: Use `agy/` prefix for Gemini models (e.g. `agy/gemini-3.5-flash-high`). Use `antigravity/` for the static curated list. Both hit the same Google backend, but expose different model naming. The quota is shared — using either provider counts against the same Google account's limits.
+**In practice**: Use `agy/` prefix for Gemini models (e.g. `agy/gemini-3.7-flash-high`). Use `antigravity/` for the static curated list. Both hit the same Google backend, but expose different model naming. The quota is shared — using either provider counts against the same Google account's limits.

 ---

--- a/open-sse/config/agyModels.ts
+++ b/open-sse/config/agyModels.ts
@@ -113,6 +113,7 @@ const AGY_RETIRED_MODEL_IDS = new Set([
  "gemini-3.6-flash-medium",
  "gemini-3.6-flash-low",
  "gemini-3-flash-agent",
+  "gemini-3.5-flash",
  "gemini-3.5-flash-extra-low",
  "gemini-3.5-flash-low",
  "gemini-3.5-flash-high",
--- a/open-sse/config/antigravityModelAliases.ts
+++ b/open-sse/config/antigravityModelAliases.ts
@@ -179,6 +179,7 @@ const ANTIGRAVITY_RETIRED_MODEL_IDS = new Set([
  "gemini-3.6-flash-medium",
  "gemini-3.6-flash-low",
  "gemini-3-flash-agent",
+  "gemini-3.5-flash",
  "gemini-3.5-flash-extra-low",
  "gemini-3.5-flash-low",
  "gemini-3.5-flash-high",
--- a/open-sse/config/geminiRateLimits.json
+++ b/open-sse/config/geminiRateLimits.json
@@ -8,7 +8,6 @@
  "gemma-4-26b-it": { "rpm": 16000, "rpd": 14400, "tpm": 16000 },
  "gemma-4-31b-it": { "rpm": 16000, "rpd": 14400, "tpm": 16000 },
  "gemini-embedding-exp-03-07": { "rpm": 100, "rpd": 1000, "tpm": 30000 },
-  "gemini-3.5-flash": { "rpm": 5, "rpd": 20, "tpm": 250000 },
  "gemini-3.1-flash-lite": { "rpm": 15, "rpd": 500, "tpm": 250000 },
  "gemini-3.1-pro": { "rpm": 0, "rpd": 0, "tpm": 0 },
  "gemini-2.5-flash-lite": { "rpm": 10, "rpd": 20, "tpm": 250000 },
--- a/open-sse/config/providers/registry/cursor/index.ts
+++ b/open-sse/config/providers/registry/cursor/index.ts
@@ -228,14 +228,14 @@ export const cursorProvider: RegistryEntry = {
    { id: "gpt-5.1-low", name: "GPT-5.1 Low" },
    { id: "gpt-5.1", name: "GPT-5.1" },
    { id: "gpt-5.1-high", name: "GPT-5.1 High" },
-    { id: "gemini-3.5-flash", name: "Gemini 3.5 Flash" },
    { id: "claude-4-sonnet", name: "Sonnet 4" },
    { id: "claude-4-sonnet-thinking", name: "Sonnet 4 Thinking" },
    { id: "gpt-5-mini", name: "GPT-5 Mini" },
    { id: "kimi-k3-low", name: "Kimi K3 Low" },
    { id: "kimi-k3-max", name: "Kimi K3" },
    { id: "glm-5.2-high", name: "GLM 5.2" },
-    { id: "glm-5.2-max", name: "GLM 5.2 Max" },  ],
+    { id: "glm-5.2-max", name: "GLM 5.2 Max" },
+  ],
 };

 /**
--- a/open-sse/services/conolModels.ts
+++ b/open-sse/services/conolModels.ts
@@ -78,7 +78,7 @@ const FALLBACK_MODEL_SEEDS: FallbackModelSeed[] = [

 /** Presets exposed by the web client's model picker (id → text/multimodal model). */
 export const CONOL_FALLBACK_MODEL_PRESETS: ConolModelPreset[] = [
-  { id: "flash", text: "deepseek/deepseek-v4-flash", multimodal: "google/gemini-3.5-flash" },
+  { id: "flash", text: "deepseek/deepseek-v4-flash", multimodal: "google/gemini-3.7-flash" },
  { id: "moderate", text: "deepseek/deepseek-v4-pro", multimodal: "claude-sonnet-5" },
  { id: "pro", text: "z-ai/glm-5.2", multimodal: "moonshotai/kimi-k3" },
  { id: "ultra", text: "claude-fable-5", multimodal: "claude-fable-5" },
--- a/open-sse/services/promptqlModels.ts
+++ b/open-sse/services/promptqlModels.ts
@@ -6,7 +6,7 @@
 */

 export interface PromptQlModel {
-  /** Client-facing id (model_reference slug, e.g. gemini-3.5-flash). */
+  /** Client-facing id (model_reference slug, e.g. gemini-3.7-flash). */
  id: string;
  /** Friendly picker label. */
  name: string;
--- a/src/lib/providerModels/geminiModelsParser.ts
+++ b/src/lib/providerModels/geminiModelsParser.ts
@@ -34,6 +34,8 @@ const IGNORED_METHODS = new Set([
  "asyncBatchEmbedContent",
 ]);

+const RETIRED_GEMINI_MODEL_IDS = new Set(["gemini-3.5-flash"]);
+
 export interface GeminiDiscoveryModel {
  id: string;
  name: string;
@@ -46,36 +48,38 @@ export interface GeminiDiscoveryModel {
 }

 export function parseGeminiModelsList(data: any): GeminiDiscoveryModel[] {
-  return (data?.models || []).map((m: Record<string, unknown>) => {
-    const methods: string[] = Array.isArray(m.supportedGenerationMethods)
-      ? (m.supportedGenerationMethods as string[])
-      : [];
+  return (data?.models || [])
+    .map((m: Record<string, unknown>) => {
+      const methods: string[] = Array.isArray(m.supportedGenerationMethods)
+        ? (m.supportedGenerationMethods as string[])
+        : [];

-    const endpoints = new Set<string>(
-      methods
-        .filter((method) => !IGNORED_METHODS.has(method))
-        .map((method) => METHOD_TO_ENDPOINT[method] || "chat")
-    );
+      const endpoints = new Set<string>(
+        methods
+          .filter((method) => !IGNORED_METHODS.has(method))
+          .map((method) => METHOD_TO_ENDPOINT[method] || "chat")
+      );

-    const id = ((m.name as string) || (m.id as string) || "").replace(/^models\//, "");
-    const lowerId = id.toLowerCase();
+      const id = ((m.name as string) || (m.id as string) || "").replace(/^models\//, "");
+      const lowerId = id.toLowerCase();

-    // Keep Veo models in the video bucket even when the method list is incomplete.
-    if (lowerId.includes("veo")) {
-      endpoints.add("video");
-    }
+      // Keep Veo models in the video bucket even when the method list is incomplete.
+      if (lowerId.includes("veo")) {
+        endpoints.add("video");
+      }

-    if (endpoints.size === 0) endpoints.add("chat");
+      if (endpoints.size === 0) endpoints.add("chat");

-    return {
-      ...m,
-      id,
-      name: (m.displayName as string) || id,
-      supportedEndpoints: [...endpoints],
-      ...(typeof m.inputTokenLimit === "number" ? { inputTokenLimit: m.inputTokenLimit } : {}),
-      ...(typeof m.outputTokenLimit === "number" ? { outputTokenLimit: m.outputTokenLimit } : {}),
-      ...(typeof m.description === "string" ? { description: m.description } : {}),
-      ...(m.thinking === true ? { supportsThinking: true } : {}),
-    } as GeminiDiscoveryModel;
-  });
+      return {
+        ...m,
+        id,
+        name: (m.displayName as string) || id,
+        supportedEndpoints: [...endpoints],
+        ...(typeof m.inputTokenLimit === "number" ? { inputTokenLimit: m.inputTokenLimit } : {}),
+        ...(typeof m.outputTokenLimit === "number" ? { outputTokenLimit: m.outputTokenLimit } : {}),
+        ...(typeof m.description === "string" ? { description: m.description } : {}),
+        ...(m.thinking === true ? { supportsThinking: true } : {}),
+      } as GeminiDiscoveryModel;
+    })
+    .filter((model: GeminiDiscoveryModel) => !RETIRED_GEMINI_MODEL_IDS.has(model.id));
 }
--- a/src/lib/providerModels/managedModelImport.ts
+++ b/src/lib/providerModels/managedModelImport.ts
@@ -20,7 +20,9 @@ import { normalizeDiscoveredModels } from "@/lib/providerModels/modelDiscovery";
 import {
  ANTIGRAVITY_MODEL_ALIASES,
  ANTIGRAVITY_REVERSE_MODEL_ALIASES,
+  isDiscoverableAntigravityModelId,
 } from "@omniroute/open-sse/config/antigravityModelAliases.ts";
+import { isDiscoverableAgyModelId } from "@omniroute/open-sse/config/agyModels.ts";
 import { filterChatSelectableModels } from "@omniroute/open-sse/services/modelEndpointPolicy.ts";
 import { filterSelectableModels } from "@omniroute/open-sse/services/modelLifecycle.ts";
 import { isSelfHostedChatProvider } from "@/shared/constants/providers";
@@ -254,14 +256,22 @@ export async function importManagedModels({
  const previousSyncedAvailableModels =
    previousSyncedAvailableModelsInput ??
    (await getSyncedAvailableModelsForConnection(providerId, connectionId));
+  const normalizedDiscoveredModels = normalizeDiscoveredModels(fetchedModels, providerId);
+  // Gemini 3.5 Flash elimination (ddf1bb760, carried from #11259): antigravity/
+  // agy discovery is restricted to each family's discoverable ids BEFORE any
+  // chat-selection filtering.
+  const providerFilteredModels =
+    providerId === "antigravity"
+      ? normalizedDiscoveredModels.filter((model) => isDiscoverableAntigravityModelId(model.id))
+      : providerId === "agy"
+        ? normalizedDiscoveredModels.filter((model) => isDiscoverableAgyModelId(model.id))
+        : normalizedDiscoveredModels;
  // #11088 (option 1): self-hosted providers keep their non-chat models — chat
  // filtering happens at read time (resolveLocalSyncedEndpointRoute). Every other
  // provider keeps the import-time chat filter: the read-time path is gated on
  // isSelfHostedChatProvider, so dropping it globally leaked image/video models
  // into OpenAI chat selections (#11271).
-  const selectableModels = filterSelectableModels(
-    providerId,
-    normalizeDiscoveredModels(fetchedModels, providerId)
+  const selectableModels = filterSelectableModels(providerId, providerFilteredModels);
  );
  const discoveredModels = isSelfHostedChatProvider(providerId)
    ? selectableModels
--- a/src/shared/constants/modelSpecs.ts
+++ b/src/shared/constants/modelSpecs.ts
@@ -99,7 +99,7 @@ const GPT_5_6_MODEL_SPEC = {
  supportsVision: true,
 } satisfies ModelSpec;

-const GEMINI_35_FLASH_MODEL_SPEC = {
+const GEMINI_36_FLASH_MODEL_SPEC = {
  maxOutputTokens: 65536,
  contextWindow: 1048576,
  supportsThinking: false,
@@ -160,7 +160,7 @@ export const MODEL_SPECS: Record<string, ModelSpec> = {
    aliases: ["openai/gpt-4o"],
  },

-  // ── Gemini 2.5 and provider-neutral 3.5 Flash series ─────────────
+  // ── Gemini 2.5 Flash ─────────────────────────────────────────────
  "gemini-2.5-flash": {
    maxOutputTokens: 65536,
    contextWindow: 1048576,
@@ -171,16 +171,6 @@ export const MODEL_SPECS: Record<string, ModelSpec> = {
    supportsTools: true,
    supportsVision: true,
  },
-  "gemini-3.5-flash-extra-low": {
-    ...GEMINI_35_FLASH_MODEL_SPEC,
-    thinkingBudgetCap: 0,
-  },
-  "gemini-3.5-flash-low": { ...GEMINI_35_FLASH_MODEL_SPEC },
-  "gemini-3-flash-agent": {
-    ...GEMINI_35_FLASH_MODEL_SPEC,
-    thinkingBudgetCap: 0,
-  },
-
  // ── Gemini 3.7 Flash (current Antigravity/AGY live tiers) ─────────
  // The tier suffix configures the thinking budget passed to the upstream
  // gemini-3.7-flash-tiered backend (high: 24.5k, medium: 8k, low: 1k).
@@ -234,9 +224,9 @@ export const MODEL_SPECS: Record<string, ModelSpec> = {
  // Provider-neutral compatibility for providers that still serve Gemini 3.6.
  // Antigravity/AGY availability is governed by their own provider catalogs and
  // retirement filters; these shared specs must not be treated as an allowlist.
-  "gemini-3.6-flash-high": { ...GEMINI_35_FLASH_MODEL_SPEC },
-  "gemini-3.6-flash-medium": { ...GEMINI_35_FLASH_MODEL_SPEC },
-  "gemini-3.6-flash-low": { ...GEMINI_35_FLASH_MODEL_SPEC },
+  "gemini-3.6-flash-high": { ...GEMINI_36_FLASH_MODEL_SPEC },
+  "gemini-3.6-flash-medium": { ...GEMINI_36_FLASH_MODEL_SPEC },
+  "gemini-3.6-flash-low": { ...GEMINI_36_FLASH_MODEL_SPEC },

  // ── Gemini 3 Flash series ───────────────────────────────────────
  "gemini-3-flash": {
@@ -282,20 +272,6 @@ export const MODEL_SPECS: Record<string, ModelSpec> = {
    aliases: ["gemini-3-pro-low"],
  },

-  // ── Gemini 3.5 Flash ─────────────────────────────────────────────
-  // #10286: the base Google AI Studio model DOES support reasoning (it has
-  // an effort-tier alias gemini-3.5-flash-high) — override the shared spec's
-  // supportsThinking:false here only. Do NOT flip GEMINI_35_FLASH_MODEL_SPEC
-  // itself: it is also spread into the Antigravity flash-tier aliases
-  // (gemini-3.5-flash-low/-extra-low, gemini-3-flash-agent, gemini-3.6-flash-*)
-  // which reject client-supplied thinking params because the model id itself
-  // selects the reasoning tier upstream.
-  "gemini-3.5-flash": {
-    ...GEMINI_35_FLASH_MODEL_SPEC,
-    supportsThinking: true,
-    aliases: ["gemini-3.5-flash-high"],
-  },
-
  // ── Claude Opus 4.5 ─────────────────────────────────────────────
  "claude-opus-4-5": {
    maxOutputTokens: 32768,
--- a/tests/unit/agy-provider.test.ts
+++ b/tests/unit/agy-provider.test.ts
@@ -57,6 +57,7 @@ test("agy ships its own live callable model catalog", () => {
  assert.ok(!ids.includes("gemini-3.6-flash-low"));
  assert.ok(!ids.includes("gemini-3.6-flash-medium"));
  assert.ok(!ids.includes("gemini-3.6-flash-high"));
+  assert.ok(!ids.includes("gemini-3.5-flash"));
  assert.ok(!ids.includes("gemini-3.5-flash-extra-low"));
  assert.ok(!ids.includes("gemini-3.5-flash-low"));
  assert.ok(!ids.includes("gemini-3-flash-agent"));
@@ -87,6 +88,7 @@ test("agy model helpers resolve catalog ids and display names", () => {
  assert.equal(isUserCallableAgyModelId("gemini-3.6-flash-low"), false);
  assert.equal(isUserCallableAgyModelId("gemini-3.6-flash-medium"), false);
  assert.equal(isUserCallableAgyModelId("gemini-3.6-flash-high"), false);
+  assert.equal(isUserCallableAgyModelId("gemini-3.5-flash"), false);
  assert.equal(isUserCallableAgyModelId("gemini-3.5-flash-extra-low"), false);
  assert.equal(isUserCallableAgyModelId("gemini-3.5-flash-low"), false);
  assert.equal(isUserCallableAgyModelId("gemini-3-flash-agent"), false);
--- a/tests/unit/antigravity-429-quota-tdd.test.ts
+++ b/tests/unit/antigravity-429-quota-tdd.test.ts
@@ -74,7 +74,7 @@ test("TDD S3: checkFallbackError extracts retry hint for oauth providers even if
    429,
    errorText,
    0,
-    "gemini-3.5-flash",
+    "gemini-3.7-flash",
    "antigravity", // which uses oauth provider profile (useUpstreamRetryHints: false)
    null
  );
--- a/tests/unit/antigravity-model-aliases.test.ts
+++ b/tests/unit/antigravity-model-aliases.test.ts
@@ -31,6 +31,7 @@ const RETIRED_FLASH_IDS = [
  "gemini-3.6-flash-low",
  "gemini-3.6-flash-medium",
  "gemini-3.6-flash-high",
+  "gemini-3.5-flash",
  "gemini-3.5-flash-extra-low",
  "gemini-3.5-flash-low",
  "gemini-3-flash-agent",
--- a/tests/unit/antigravity-retired-public-models.test.ts
+++ b/tests/unit/antigravity-retired-public-models.test.ts
@@ -21,6 +21,7 @@ const RETIRED_PUBLIC_MODELS = [
  "gemini-3.6-flash-medium",
  "gemini-3.6-flash-low",
  "gemini-3-flash-agent",
+  "gemini-3.5-flash",
  "gemini-3.5-flash-low",
  "gemini-3.5-flash-extra-low",
  "gemini-2.5-pro",
--- a/tests/unit/conol-web.test.ts
+++ b/tests/unit/conol-web.test.ts
@@ -14,6 +14,7 @@ import {
 } from "../../open-sse/executors/conol-web.ts";
 import {
  CONOL_FALLBACK_MODELS,
+  CONOL_FALLBACK_MODEL_PRESETS,
  clampConolEffort,
  parseConolAgentServers,
  resolveConolModelSelection,
@@ -27,6 +28,11 @@ import { getResolvedModelCapabilities } from "../../src/lib/modelCapabilities.ts
 const SESSION_COOKIE_NAME = "__Secure-better-auth.session_token";

 describe("Conol web provider", () => {
+  it("routes the Flash preset multimodal path to Gemini 3.7", () => {
+    const flashPreset = CONOL_FALLBACK_MODEL_PRESETS.find((preset) => preset.id === "flash");
+    assert.equal(flashPreset?.multimodal, "google/gemini-3.7-flash");
+  });
+
  it("normalizes raw, full-header, JSON, and provider-data credentials", () => {
    assert.equal(normalizeConolCookie("token-value"), `${SESSION_COOKIE_NAME}=token-value`);
    assert.equal(
--- a/tests/unit/copilot-gemini-claude-route-no-responses.test.ts
+++ b/tests/unit/copilot-gemini-claude-route-no-responses.test.ts
@@ -55,7 +55,7 @@ describe("GithubExecutor — Gemini/Claude must never hit /responses (port 9rout

  it("routes registered Gemini Copilot models to chat/completions", () => {
    const exec = new GithubExecutor();
-    for (const id of ["gemini-3.1-pro-preview", "gemini-3.5-flash"]) {
+    for (const id of ["gemini-3.1-pro-preview", "gemini-3.7-flash"]) {
      assert.equal(exec.buildUrl(id, false), CHAT_URL, `${id} must route to chat/completions`);
    }
  });
--- a/tests/unit/cursor-registry-claude-families.test.ts
+++ b/tests/unit/cursor-registry-claude-families.test.ts
@@ -8,6 +8,10 @@ function modelIds(): Set<string> {
  return new Set(cursorProvider.models.map((m) => m.id));
 }

+test("cursor registry excludes retired Gemini 3.5 Flash", () => {
+  assert.equal(modelIds().has("gemini-3.5-flash"), false);
+});
+
 test("cursor registry includes Claude Opus 4.8 effort + thinking + fast variants", () => {
  const ids = modelIds();
  for (const effort of EFFORTS) {
--- a/tests/unit/executor-promptql.test.ts
+++ b/tests/unit/executor-promptql.test.ts
@@ -56,7 +56,7 @@ describe("PromptQl — registry consistency", () => {
  it("registers a model catalog via getModelsByProviderId", () => {
    const catalog = getModelsByProviderId("promptql");
    assert.ok(catalog.length >= 5);
-    assert.ok(catalog.some((m) => m.id === "gemini-3.5-flash" || m.id.includes("gemini")));
+    assert.ok(catalog.some((m) => m.id === "gemini-3.7-flash" || m.id.includes("gemini")));
    assert.ok(catalog.some((m) => m.id.includes("gpt-5.6") || m.id.includes("fable")));
  });
 });
@@ -209,7 +209,10 @@ describe("PromptQl — helpers", () => {
  });

  it("resolves model slugs and prefixes", () => {
-    assert.equal(models.clientFacingPromptQlModelId("promptql/gemini-3.5-flash"), "gemini-3.5-flash");
+    assert.equal(
+      models.clientFacingPromptQlModelId("promptql/gemini-3.7-flash"),
+      "gemini-3.7-flash"
+    );
    assert.equal(models.clientFacingPromptQlModelId("pql/gpt-5.6-sol"), "gpt-5.6-sol");
    const r = models.resolvePromptQlModel("Claude Fable 5");
    assert.ok(r);
@@ -265,7 +268,7 @@ describe("PromptQlExecutor — auth / validation", () => {
  it("returns 401 when no token is supplied", async () => {
    const executor = new mod.PromptQlExecutor();
    const result = await executor.execute({
-      model: "gemini-3.5-flash",
+      model: "gemini-3.7-flash",
      body: { messages: [{ role: "user", content: "hi" }] },
      stream: false,
      credentials: {},
@@ -279,7 +282,7 @@ describe("PromptQlExecutor — auth / validation", () => {
  it("returns 400 when no user message is present", async () => {
    const executor = new mod.PromptQlExecutor();
    const result = await executor.execute({
-      model: "gemini-3.5-flash",
+      model: "gemini-3.7-flash",
      body: { messages: [{ role: "assistant", content: "hi" }] },
      stream: false,
      credentials: { apiKey: sampleJwt },
@@ -614,7 +617,7 @@ describe("PromptQlExecutor — mocked GraphQL turn", () => {
    try {
      const executor = new mod.PromptQlExecutor();
      const result = await executor.execute({
-        model: "gemini-3.5-flash",
+        model: "gemini-3.7-flash",
        body: { messages: [{ role: "user", content: "ping" }] },
        stream: false,
        credentials: { apiKey: sampleJwt },
@@ -628,7 +631,7 @@ describe("PromptQlExecutor — mocked GraphQL turn", () => {
      };
      assert.equal(json.choices[0]!.message.content, "HELLO-PQL");
      assert.equal(json.promptql_thread_id, "thread-1");
-      assert.equal(json.model, "gemini-3.5-flash");
+      assert.equal(json.model, "gemini-3.7-flash");
      assert.ok(call >= 2);
      assert.equal(result.response.headers.get("X-PromptQL-Thread-Id"), "thread-1");
    } finally {
--- a/tests/unit/gemini-3-5-flash-thinking.test.ts
+++ b/tests/unit/gemini-3-5-flash-thinking.test.ts
@@ -1,76 +0,0 @@
-// Regression test for #10286: gemini-3.5-flash was incorrectly marked
-// supportsThinking:false, causing a spurious pre-provider HTTP 400 for any
-// request with reasoning_effort set, even though the base Google AI Studio
-// model supports reasoning (it has an effort-tier alias gemini-3.5-flash-high).
-import test from "node:test";
-import assert from "node:assert/strict";
-import fs from "node:fs";
-import os from "node:os";
-import path from "node:path";
-
-const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-repro-10286-"));
-process.env.DATA_DIR = TEST_DATA_DIR;
-process.env.API_KEY_SECRET = process.env.API_KEY_SECRET || "test-repro-10286-secret";
-
-const caps = await import("../../src/lib/modelCapabilities.ts");
-const core = await import("../../src/lib/db/core.ts");
-const rulesDb = await import("../../src/lib/db/reasoningRoutingRules.ts");
-const policy = await import("../../src/lib/reasoningRouting/policy.ts");
-
-async function resetStorage() {
-  core.resetDbInstance();
-  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
-  fs.mkdirSync(TEST_DATA_DIR, { recursive: true });
-  rulesDb.invalidateReasoningRoutingRuleCache();
-}
-
-function ruleInput(patch: Record<string, unknown> = {}) {
-  return {
-    name: "Enable thinking on gemini-3.5-flash",
-    description: "",
-    scope: "global",
-    apiKeyId: null,
-    comboId: null,
-    connectionId: null,
-    modelPattern: "gemini-3.5-flash",
-    sourceEffort: "any",
-    requestTags: [],
-    tagMatchMode: "any",
-    effortMode: "inherit",
-    targetEffort: null,
-    targetKind: "keep",
-    targetModel: null,
-    targetComboId: null,
-    budgetAction: "preserve",
-    budgetTokens: null,
-    priority: 0,
-    enabled: true,
-    ...patch,
-  };
-}
-
-test.beforeEach(resetStorage);
-test.after(async () => {
-  await resetStorage();
-  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
-});
-
-test("gemini-3.5-flash (AI Studio provider) resolves as thinking-capable", () => {
-  const resolved = caps.getResolvedModelCapabilities({
-    provider: "gemini",
-    model: "gemini-3.5-flash",
-  });
-  assert.equal(resolved.supportsThinking, true);
-});
-
-test("reasoning_effort 'high' on gemini-3.5-flash is NOT rejected by routing policy", async () => {
-  await rulesDb.createReasoningRoutingRule(ruleInput());
-  const decision = await policy.resolveReasoningRoutingRule({
-    sourceModel: "gemini/gemini-3.5-flash",
-    sourceModelAliases: ["gemini-3.5-flash"],
-    sourceEffort: "high",
-    hasReasoningSignal: true,
-  });
-  assert.ok(decision, "a matching rule must produce a decision");
-  assert.equal(decision.capability, "supported");
-});
--- a/tests/unit/gemini-codex-encrypted-tool-schema.test.ts
+++ b/tests/unit/gemini-codex-encrypted-tool-schema.test.ts
@@ -71,7 +71,7 @@ test("OpenAI -> Gemini request strips encrypted from Codex collaboration tool pa
    ],
  };

-  const result = openaiToGeminiRequest("gemini-3.5-flash-low", body, false) as {
+  const result = openaiToGeminiRequest("gemini-3.7-flash-low", body, false) as {
    tools?: Array<{ functionDeclarations?: Array<{ parameters: unknown }> }>;
  };

--- a/tests/unit/gemini-models-parser.test.ts
+++ b/tests/unit/gemini-models-parser.test.ts
@@ -14,6 +14,16 @@ const SAMPLE = {
      supportedGenerationMethods: ["generateContent", "countTokens", "batchGenerateContent"],
      thinking: true,
    },
+    {
+      name: "models/gemini-3.5-flash",
+      displayName: "Gemini 3.5 Flash",
+      supportedGenerationMethods: ["generateContent"],
+    },
+    {
+      name: "models/gemini-3.5-flash-lite",
+      displayName: "Gemini 3.5 Flash Lite",
+      supportedGenerationMethods: ["generateContent"],
+    },
    {
      name: "models/gemini-3-pro-image-preview",
      displayName: "Gemini 3 Pro Image Preview",
@@ -48,6 +58,12 @@ test("parseGeminiModelsList strips the models/ prefix and maps display name", ()
  assert.deepEqual(flash!.supportedEndpoints, ["chat"]);
 });

+test("parseGeminiModelsList excludes retired Gemini 3.5 Flash but keeps Flash Lite", () => {
+  const ids = parseGeminiModelsList(SAMPLE).map((model) => model.id);
+  assert.equal(ids.includes("gemini-3.5-flash"), false);
+  assert.equal(ids.includes("gemini-3.5-flash-lite"), true);
+});
+
 test("parseGeminiModelsList maps generateContent image models to the chat endpoint", () => {
  const models = parseGeminiModelsList(SAMPLE);
  const proImage = models.find((m) => m.id === "gemini-3-pro-image-preview");
--- a/tests/unit/gemini-strict-tool-schema.test.ts
+++ b/tests/unit/gemini-strict-tool-schema.test.ts
@@ -56,7 +56,7 @@ test("OpenAI -> Gemini request strips strict from OpenAI-style function tool par
    ],
  };

-  const result = openaiToGeminiRequest("gemini-3.5-flash-low", body, false) as {
+  const result = openaiToGeminiRequest("gemini-3.7-flash-low", body, false) as {
    tools?: Array<{ functionDeclarations?: Array<{ parameters: unknown }> }>;
  };

--- a/tests/unit/ghe-copilot.test.ts
+++ b/tests/unit/ghe-copilot.test.ts
@@ -99,7 +99,7 @@ test("buildUrl uses chat/completions endpoint for gemini models", () => {
  };
  // Gemini has no native shim on Copilot — it stays on /chat/completions.
  assert.strictEqual(
-    executor.buildUrl("gemini-3.5-flash", true, 0, credentials),
+    executor.buildUrl("gemini-3.7-flash", true, 0, credentials),
    "https://ghe.company.com/chat/completions"
  );
 });
--- a/tests/unit/managed-model-import.test.ts
+++ b/tests/unit/managed-model-import.test.ts
@@ -279,6 +279,7 @@ test("antigravity sync dynamically builds and saves mitmAlias mappings", async (
    mode: "sync",
    fetchedModels: [
      { id: "gemini-3.5-flash", name: "Gemini 3.5 Flash" },
+      { id: "gemini-3.7-flash-high", name: "Gemini 3.7 Flash High" },
      { id: "custom-antigravity-model", name: "Custom Antigravity Model" },
    ],
  });
@@ -289,8 +290,13 @@ test("antigravity sync dynamically builds and saves mitmAlias mappings", async (
  const mitmMappings = await modelsDb.getMitmAlias("antigravity");
  console.log("MITM MAPPINGS IN TEST:", mitmMappings);

-  // Should contain standard mapping
-  assert.equal(mitmMappings["gemini-3.5-flash"], "antigravity/gemini-3.5-flash");
+  // Retired models reported by upstream must not be imported or mapped.
+  assert.equal(mitmMappings["gemini-3.5-flash"], undefined);
+  assert.equal(
+    models.some((model) => model.id === "gemini-3.5-flash"),
+    false
+  );
+  assert.equal(mitmMappings["gemini-3.7-flash-high"], "antigravity/gemini-3.7-flash-high");
  assert.equal(mitmMappings["custom-antigravity-model"], "antigravity/custom-antigravity-model");

  // Removed Antigravity 2.0 preview/agent aliases must not be reintroduced.
--- a/tests/unit/model-capabilities-registry.test.ts
+++ b/tests/unit/model-capabilities-registry.test.ts
@@ -154,21 +154,14 @@ test("unknown models keep maxOutputTokens null instead of using a generic defaul
  );
 });

-test("provider-neutral Gemini 3.5 tier IDs retain their non-thinking capabilities", () => {
+test("retired Gemini 3.5 Flash IDs have no provider-neutral model specs", () => {
  for (const modelId of [
+    "gemini-3.5-flash",
    "gemini-3.5-flash-extra-low",
    "gemini-3.5-flash-low",
    "gemini-3-flash-agent",
  ]) {
-    const spec = MODEL_SPECS[modelId];
-    assert.ok(spec, `missing exact MODEL_SPECS entry for ${modelId}`);
-    const capabilities = modelCapabilities.getResolvedModelCapabilities(modelId);
-    assert.equal(capabilities.contextWindow, 1048576, modelId);
-    assert.equal(capabilities.maxOutputTokens, 65536, modelId);
-    // These ids encode the upstream reasoning tier and do not accept a client-supplied effort.
-    assert.equal(capabilities.supportsThinking, false, modelId);
-    assert.equal(capabilities.supportsTools, true, modelId);
-    assert.equal(capabilities.supportsVision, true, modelId);
+    assert.equal(MODEL_SPECS[modelId], undefined, modelId);
  }
 });

--- a/tests/unit/models-catalog-route.test.ts
+++ b/tests/unit/models-catalog-route.test.ts
@@ -702,6 +702,7 @@ test("v1 models catalog exposes current Antigravity aliases without retired mode
  assert.equal(ids.has("antigravity/gemini-3.6-flash-high"), false);
  assert.equal(ids.has("antigravity/gemini-3.6-flash-medium"), false);
  assert.equal(ids.has("antigravity/gemini-3.6-flash-low"), false);
+  assert.equal(ids.has("antigravity/gemini-3.5-flash"), false);
  assert.equal(ids.has("antigravity/gemini-3.5-flash-extra-low"), false);
  assert.equal(ids.has("antigravity/gemini-3.5-flash-low"), false);
  assert.equal(ids.has("antigravity/gemini-3-flash-agent"), false);
--- a/tests/unit/omniroute-response-meta.test.ts
+++ b/tests/unit/omniroute-response-meta.test.ts
@@ -60,7 +60,7 @@ test("buildOmniRouteResponseMetaHeaders keeps ASCII model header values unchange
 });

 test("buildOmniRouteResponseMetaHeaders percent-encodes non-ASCII model header values", () => {
-  const model = "free-mix/[假流式]gemini-3.5-flash";
+  const model = "free-mix/[假流式]gemini-3.7-flash";
  const headers = buildOmniRouteResponseMetaHeaders({
    provider: "openai",
    model,
--- a/tests/unit/provider-limits-ui.test.ts
+++ b/tests/unit/provider-limits-ui.test.ts
@@ -288,7 +288,7 @@ test("hidden provider models are filtered from per-model quota rows", () => {
  });
  const hidden = providerLimitUtils.collectHiddenQuotaModelIds("antigravity", {
    models: [{ id: "antigravity/gpt-oss-120b-medium", isHidden: true }],
-    modelCompatOverrides: [{ id: "gemini-3.5-flash", isHidden: true }],
+    modelCompatOverrides: [{ id: "gemini-3.7-flash", isHidden: true }],
  });
  const visible = providerLimitUtils.filterHiddenModelQuotas("antigravity", quotas, hidden);

--- a/tests/unit/sse-parser.test.ts
+++ b/tests/unit/sse-parser.test.ts
@@ -394,7 +394,7 @@ test("parseSSEToGeminiResponse extracts tool calls from textual format", () => {
    })}`,
  ].join("\n");

-  const parsed = parseSSEToGeminiResponse(rawSSE, "gemini-3.5-flash-low");
+  const parsed = parseSSEToGeminiResponse(rawSSE, "gemini-3.7-flash-low");

  assert.ok(parsed);
  assert.equal(parsed.choices[0].finish_reason, "tool_calls");
--- a/tests/unit/stream-utils.test.ts
+++ b/tests/unit/stream-utils.test.ts
@@ -232,14 +232,14 @@ test("createSSEStream passthrough converts textual tool-call content into struct
        id: "chatcmpl_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: { role: "assistant", content: toolText } }],
      })}\n\n`,
      `data: ${JSON.stringify({
        id: "chatcmpl_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
      })}\n\n`,
    ],
@@ -247,7 +247,7 @@ test("createSSEStream passthrough converts textual tool-call content into struct
      mode: "passthrough",
      sourceFormat: FORMATS.OPENAI,
      provider: "antigravity",
-      model: "antigravity/gemini-3.5-flash-low",
+      model: "antigravity/gemini-3.7-flash-low",
      body: {
        messages: [{ role: "user", content: "inspect db" }],
      },
@@ -284,21 +284,21 @@ test("createSSEStream passthrough converts split textual tool-call content at co
        id: "chatcmpl_split_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: { role: "assistant", content: chunks[0] } }],
      })}\n\n`,
      `data: ${JSON.stringify({
        id: "chatcmpl_split_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: { content: chunks[1] } }],
      })}\n\n`,
      `data: ${JSON.stringify({
        id: "chatcmpl_split_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
      })}\n\n`,
    ],
@@ -306,7 +306,7 @@ test("createSSEStream passthrough converts split textual tool-call content at co
      mode: "passthrough",
      sourceFormat: FORMATS.OPENAI,
      provider: "antigravity",
-      model: "antigravity/gemini-3.5-flash-low",
+      model: "antigravity/gemini-3.7-flash-low",
      body: { messages: [{ role: "user", content: "inspect db" }] },
      onComplete(payload) {
        onCompletePayload = payload;
@@ -340,28 +340,28 @@ test("createSSEStream passthrough handles textual tool-call content split inside
        id: "chatcmpl_split_prefix_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: { role: "assistant", content: chunks[0] } }],
      })}\n\n`,
      `data: ${JSON.stringify({
        id: "chatcmpl_split_prefix_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: { content: chunks[1] } }],
      })}\n\n`,
      `data: ${JSON.stringify({
        id: "chatcmpl_split_prefix_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: { content: chunks[2] } }],
      })}\n\n`,
      `data: ${JSON.stringify({
        id: "chatcmpl_split_prefix_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
      })}\n\n`,
    ],
@@ -369,7 +369,7 @@ test("createSSEStream passthrough handles textual tool-call content split inside
      mode: "passthrough",
      sourceFormat: FORMATS.OPENAI,
      provider: "antigravity",
-      model: "antigravity/gemini-3.5-flash-low",
+      model: "antigravity/gemini-3.7-flash-low",
      body: { messages: [{ role: "user", content: "inspect db" }] },
      onComplete(payload) {
        onCompletePayload = payload;
@@ -515,14 +515,14 @@ Arguments: {"path":"/opt/OmniRoute/src","target":"files"}`;
        id: "chatcmpl_unknown_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: { role: "assistant", content: toolText } }],
      })}\n\n`,
      `data: ${JSON.stringify({
        id: "chatcmpl_unknown_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
      })}\n\n`,
    ],
@@ -530,7 +530,7 @@ Arguments: {"path":"/opt/OmniRoute/src","target":"files"}`;
      mode: "passthrough",
      sourceFormat: FORMATS.OPENAI,
      provider: "antigravity",
-      model: "antigravity/gemini-3.5-flash-low",
+      model: "antigravity/gemini-3.7-flash-low",
      body: {
        messages: [{ role: "user", content: "inspect files" }],
        tools: [
@@ -561,14 +561,14 @@ test("createSSEStream passthrough suppresses malformed textual tool-call content
        id: "chatcmpl_malformed_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: { role: "assistant", content: malformedToolText } }],
      })}\n\n`,
      `data: ${JSON.stringify({
        id: "chatcmpl_malformed_textual_tool",
        object: "chat.completion.chunk",
        created: 1,
-        model: "antigravity/gemini-3.5-flash-low",
+        model: "antigravity/gemini-3.7-flash-low",
        choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
      })}\n\n`,
    ],
@@ -576,7 +576,7 @@ test("createSSEStream passthrough suppresses malformed textual tool-call content
      mode: "passthrough",
      sourceFormat: FORMATS.OPENAI,
      provider: "antigravity",
-      model: "antigravity/gemini-3.5-flash-low",
+      model: "antigravity/gemini-3.7-flash-low",
      body: { messages: [{ role: "user", content: "inspect db" }] },
      onComplete(payload) {
        onCompletePayload = payload;
@@ -617,7 +617,7 @@ test("createSSEStream suppresses malformed compact textual tool-call content", a
      targetFormat: FORMATS.ANTIGRAVITY,
      sourceFormat: FORMATS.OPENAI,
      provider: "antigravity",
-      model: "antigravity/gemini-3.5-flash-low",
+      model: "antigravity/gemini-3.7-flash-low",
      body: { messages: [{ role: "user", content: "inspect files" }] },
      onComplete(payload) {
        onCompletePayload = payload;
@@ -1024,7 +1024,7 @@ Arguments: {"command":"systemctl status omniroute"}`;
        response: {
          id: "resp_textual_tool",
          object: "response",
-          model: "antigravity/gemini-3.5-flash-low",
+          model: "antigravity/gemini-3.7-flash-low",
          status: "completed",
          output: [],
          usage: { input_tokens: 10, output_tokens: 4, total_tokens: 14 },
@@ -1038,7 +1038,7 @@ Arguments: {"command":"systemctl status omniroute"}`;
      sourceFormat: FORMATS.OPENAI_RESPONSES,
      clientResponseFormat: FORMATS.OPENAI_RESPONSES,
      provider: "antigravity",
-      model: "antigravity/gemini-3.5-flash-low",
+      model: "antigravity/gemini-3.7-flash-low",
      body: {
        input: "check service",
        tools: [{ type: "function", name: "terminal", parameters: { type: "object" } }],
--- a/tests/unit/t28-model-catalog-updates.test.ts
+++ b/tests/unit/t28-model-catalog-updates.test.ts
@@ -30,6 +30,7 @@ test("T28: antigravity static catalog exposes only callable Gemini tier IDs", ()
  assert.ok(!staticIds.includes("gemini-3.6-flash-high"));
  assert.ok(!staticIds.includes("gemini-3.6-flash-medium"));
  assert.ok(!staticIds.includes("gemini-3.6-flash-low"));
+  assert.ok(!staticIds.includes("gemini-3.5-flash"));
  assert.ok(!staticIds.includes("gemini-3.5-flash-extra-low"));
  assert.ok(!staticIds.includes("gemini-3.5-flash-low"));
  assert.ok(!staticIds.includes("gemini-3-flash-agent"));
--- a/tests/unit/translator-openai-to-gemini.test.ts
+++ b/tests/unit/translator-openai-to-gemini.test.ts
@@ -607,7 +607,7 @@ test("OpenAI -> Antigravity wraps Gemini requests in a Cloud Code envelope", ()

 test("OpenAI -> Antigravity Gemini omits signature-less historical tool calls and keeps response context", () => {
  const result = openaiToAntigravityRequest(
-    "gemini-3.5-flash-low",
+    "gemini-3.7-flash-low",
    {
      messages: [
        { role: "user", content: "Update todo" },
@@ -686,7 +686,7 @@ test("OpenAI -> Antigravity Gemini omits signature-less historical tool calls an

 test("OpenAI -> Antigravity preserves multiple signature-less historical tool responses as context", () => {
  const result = openaiToAntigravityRequest(
-    "gemini-3.5-flash-low",
+    "gemini-3.7-flash-low",
    {
      messages: [
        { role: "user", content: "Inspect OmniRoute config" },
@@ -747,7 +747,7 @@ test("OpenAI -> Antigravity preserves signed Gemini tool calls in native form",
  storeGeminiThoughtSignature(buildGeminiThoughtSignatureKey(ns, toolId), "SIG_AG_SIGNED_XYZ");

  const result = openaiToAntigravityRequest(
-    "gemini-3.5-flash-low",
+    "gemini-3.7-flash-low",
    {
      messages: [
        { role: "user", content: "Read status" },
@@ -787,7 +787,7 @@ test("OpenAI -> Antigravity preserves signed Gemini tool calls in native form",

 test("OpenAI -> Antigravity escapes signature-less tool response context content", () => {
  const result = openaiToAntigravityRequest(
-    "gemini-3.5-flash-low",
+    "gemini-3.7-flash-low",
    {
      messages: [
        { role: "user", content: "Inspect previous output" },
--- a/tests/unit/translator-resp-gemini-to-openai.test.ts
+++ b/tests/unit/translator-resp-gemini-to-openai.test.ts
@@ -346,7 +346,7 @@ test("Gemini stream: converts textual Tool call block to structured tool_calls",
  const result = geminiToOpenAIResponse(
    {
      responseId: "resp-textual-tool",
-      modelVersion: "gemini-3.5-flash-low",
+      modelVersion: "gemini-3.7-flash-low",
      candidates: [
        {
          content: {
@@ -386,7 +386,7 @@ test("Gemini stream: routes textual reasoning tags to reasoning_content before t
  const result = geminiToOpenAIResponse(
    {
      responseId: "resp-textual-thought-tool",
-      modelVersion: "gemini-3.5-flash-high",
+      modelVersion: "gemini-3.7-flash-high",
      candidates: [
        {
          content: {
@@ -431,7 +431,7 @@ test("Gemini stream: keeps textual reasoning hidden across split chunks", () =>
  const first = geminiToOpenAIResponse(
    {
      responseId: "resp-split-thought",
-      modelVersion: "gemini-3.5-flash-high",
+      modelVersion: "gemini-3.7-flash-high",
      candidates: [{ content: { parts: [{ text: "§54§ <tho" }] } }],
    },
    state
@@ -444,7 +444,7 @@ test("Gemini stream: keeps textual reasoning hidden across split chunks", () =>
  const second = geminiToOpenAIResponse(
    {
      responseId: "resp-split-thought",
-      modelVersion: "gemini-3.5-flash-high",
+      modelVersion: "gemini-3.7-flash-high",
      candidates: [{ content: { parts: [{ text: "ught\nNeed to inspect" }] } }],
    },
    state
@@ -459,7 +459,7 @@ test("Gemini stream: keeps textual reasoning hidden across split chunks", () =>
  const third = geminiToOpenAIResponse(
    {
      responseId: "resp-split-thought",
-      modelVersion: "gemini-3.5-flash-high",
+      modelVersion: "gemini-3.7-flash-high",
      candidates: [{ content: { parts: [{ text: " more</tho" }] } }],
    },
    state
@@ -472,7 +472,7 @@ test("Gemini stream: keeps textual reasoning hidden across split chunks", () =>
  const fourth = geminiToOpenAIResponse(
    {
      responseId: "resp-split-thought",
-      modelVersion: "gemini-3.5-flash-high",
+      modelVersion: "gemini-3.7-flash-high",
      candidates: [{ content: { parts: [{ text: "ught>Visible answer" }] } }],
    },
    state
@@ -498,7 +498,7 @@ test("Gemini stream: converts prefixed textual Tool call block with zero-width c
  const result = geminiToOpenAIResponse(
    {
      responseId: "resp-textual-tool-prefixed",
-      modelVersion: "gemini-3.5-flash-low",
+      modelVersion: "gemini-3.7-flash-low",
      candidates: [
        {
          content: {
@@ -612,7 +612,7 @@ test("Gemini stream: unwraps native functionCall args when emitted as JSON strin
  const result = geminiToOpenAIResponse(
    {
      responseId: "resp-native-tool-json-string",
-      modelVersion: "gemini-3.5-flash-low",
+      modelVersion: "gemini-3.7-flash-low",
      candidates: [
        {
          content: {
@@ -648,7 +648,7 @@ test("Gemini stream: converts JSON-string encoded textual Tool call arguments",
  const result = geminiToOpenAIResponse(
    {
      responseId: "resp-textual-tool-json-string",
-      modelVersion: "gemini-3.5-flash-low",
+      modelVersion: "gemini-3.7-flash-low",
      candidates: [
        {
          content: {
@@ -685,7 +685,7 @@ test("Gemini stream: suppresses malformed textual Tool call marker", () => {
  const result = geminiToOpenAIResponse(
    {
      responseId: "resp-textual-tool-malformed",
-      modelVersion: "gemini-3.5-flash-low",
+      modelVersion: "gemini-3.7-flash-low",
      candidates: [
        {
          content: {
@@ -717,7 +717,7 @@ test("Gemini stream: handles textual Tool call block split across chunks", () =>
  const state = createStreamingState();
  const chunk1 = {
    responseId: "resp-split",
-    modelVersion: "gemini-3.5-flash-low",
+    modelVersion: "gemini-3.7-flash-low",
    candidates: [
      {
        content: {
@@ -732,7 +732,7 @@ test("Gemini stream: handles textual Tool call block split across chunks", () =>
  };
  const chunk2 = {
    responseId: "resp-split",
-    modelVersion: "gemini-3.5-flash-low",
+    modelVersion: "gemini-3.7-flash-low",
    candidates: [
      {
        content: {
@@ -768,7 +768,7 @@ test("Gemini stream: does not swallow false positive textual tool call in backti
  const state = createStreamingState();
  const chunk1 = {
    responseId: "resp-false-positive",
-    modelVersion: "gemini-3.5-flash-low",
+    modelVersion: "gemini-3.7-flash-low",
    candidates: [
      {
        content: {
@@ -783,7 +783,7 @@ test("Gemini stream: does not swallow false positive textual tool call in backti
  };
  const chunk2 = {
    responseId: "resp-false-positive",
-    modelVersion: "gemini-3.5-flash-low",
+    modelVersion: "gemini-3.7-flash-low",
    candidates: [
      {
        content: {
@@ -817,7 +817,7 @@ test("Gemini stream: does not swallow terminated trailing false positive textual
  const state = createStreamingState();
  const chunk1 = {
    responseId: "resp-false-positive-terminated",
-    modelVersion: "gemini-3.5-flash-low",
+    modelVersion: "gemini-3.7-flash-low",
    candidates: [
      {
        content: {
@@ -842,7 +842,7 @@ test("Gemini stream: flushes left part before textual tool call candidate and fl
  const state = createStreamingState() as any;
  const chunk1 = {
    responseId: "resp-test-flush-left",
-    modelVersion: "gemini-3.5-flash-low",
+    modelVersion: "gemini-3.7-flash-low",
    candidates: [
      {
        content: {
@@ -884,7 +884,7 @@ test("Gemini stream: splits mid-stream partial candidate but preserves tool call
  const state = createStreamingState() as any;
  const chunk1 = {
    responseId: "resp-test-split-candidate",
-    modelVersion: "gemini-3.5-flash-low",
+    modelVersion: "gemini-3.7-flash-low",
    candidates: [
      {
        content: {
@@ -932,7 +932,7 @@ test("Gemini stream: index mismatch regression test with zero-width characters i
  const result = geminiToOpenAIResponse(
    {
      responseId: "resp-textual-tool-index-mismatch",
-      modelVersion: "gemini-3.5-flash-low",
+      modelVersion: "gemini-3.7-flash-low",
      candidates: [
        {
          content: {
@@ -964,7 +964,7 @@ test("Gemini stream: partial tool call with (empty) prefix check at chunk end do
  const state = createStreamingState();
  const chunk1 = {
    responseId: "resp-empty-leak",
-    modelVersion: "gemini-3.5-flash-low",
+    modelVersion: "gemini-3.7-flash-low",
    candidates: [
      {
        content: {
@@ -1011,7 +1011,7 @@ test("Gemini stream: parses textual tool call that starts in a subsequent chunk
  const state = createStreamingState() as any;
  const chunk1 = {
    responseId: "resp-test-after-prose",
-    modelVersion: "gemini-3.5-flash-low",
+    modelVersion: "gemini-3.7-flash-low",
    candidates: [
      {
        content: {
@@ -1060,7 +1060,7 @@ test("Gemini stream: checks lastParen before lastBracket when identifying partia
  // Имитируем чанк, который кончается на частичный "(empty)[Tool call:" маркер, например "(em"
  const chunk1 = {
    responseId: "resp-test-empty-partial",
-    modelVersion: "gemini-3.5-flash-low",
+    modelVersion: "gemini-3.7-flash-low",
    candidates: [
      {
        content: {
@@ -1187,7 +1187,7 @@ test("Gemini stream: partial textual tool call survives a reasoning-only chunk",
  geminiToOpenAIResponse(
    {
      responseId: "resp-interleave",
-      modelVersion: "gemini-3.5-flash-low",
+      modelVersion: "gemini-3.7-flash-low",
      candidates: [
        { content: { parts: [{ text: '[Tool call: terminal]\nArguments: {"command":"ls' }] } },
      ],
@@ -1200,7 +1200,7 @@ test("Gemini stream: partial textual tool call survives a reasoning-only chunk",
    geminiToOpenAIResponse(
      {
        responseId: "resp-interleave",
-        modelVersion: "gemini-3.5-flash-low",
+        modelVersion: "gemini-3.7-flash-low",
        candidates: [{ content: { parts: [{ text: "<thinking>pondering</thinking>" }] } }],
      },
      state
@@ -1221,7 +1221,7 @@ test("Gemini stream: partial textual tool call survives a reasoning-only chunk",
    geminiToOpenAIResponse(
      {
        responseId: "resp-interleave",
-        modelVersion: "gemini-3.5-flash-low",
+        modelVersion: "gemini-3.7-flash-low",
        candidates: [{ content: { parts: [{ text: '"}' }] }, finishReason: "STOP" }],
      },
      state
				`@@ -1 +0,0 @@`
				`- fix(sse): mark gemini-3.5-flash as thinking-capable so reasoning_effort is no longer rejected with a spurious 400 (#10286)`