diff --git a/packages/opencode/src/provider/transform.ts b/packages/opencode/src/provider/transform.ts
index 7602d92af5..9ea2cbc556 100644
--- a/packages/opencode/src/provider/transform.ts
+++ b/packages/opencode/src/provider/transform.ts
@@ -1347,7 +1347,7 @@ export namespace ProviderTransform {
     ) {
       return mergeDeep({ store: false }, small)
     }
-    // altimate_change start — upstream_fix: restore Google minimal thinking for small calls only on reasoning models
+    // altimate_change start — upstream_fix: use generation-supported Google thinking controls for small reasoning calls
     if (
       (model.providerID === "google" ||
         model.api.npm === "@ai-sdk/google" ||
@@ -1355,10 +1355,19 @@ export namespace ProviderTransform {
       model.capabilities.reasoning
     ) {
       const id = model.api.id.toLowerCase()
-      if (id.includes("gemini-3")) {
+      // Leading digits after "gemini-", wherever it appears (prefixed, concatenated or dated ids).
+      const generation = Number(/gemini-(\d+)/.exec(id)?.[1])
+      if (generation === 3) {
         return { thinkingConfig: { thinkingLevel: "minimal" } }
       }
-      return { thinkingConfig: { thinkingBudget: 0 } }
+      // Gemini 2.x and unversioned aliases (e.g. gemini-flash-latest) keep the budget control.
+      if (!(generation >= 4)) {
+        return { thinkingConfig: { thinkingBudget: 0 } }
+      }
+      // Gemini 4+ rejects budgets and "minimal"; prefer the lowest configured supported level.
+      const levels = Object.values(model.variants ?? {}).map((variant) => variant.thinkingConfig?.thinkingLevel)
+      const thinkingLevel = WIDELY_SUPPORTED_EFFORTS.find((level) => levels.includes(level)) ?? "low"
+      return { thinkingConfig: { thinkingLevel } }
     }
     // altimate_change end
     if (model.providerID === "openrouter" || model.providerID === "llmgateway") {
diff --git a/packages/opencode/test/provider/transform.test.ts b/packages/opencode/test/provider/transform.test.ts
index fb9ed195c9..f5ac6420c2 100644
--- a/packages/opencode/test/provider/transform.test.ts
+++ b/packages/opencode/test/provider/transform.test.ts
@@ -307,6 +307,55 @@ describe("ProviderTransform.smallOptions - google thinking", () => {
       variants: overrides.variants ?? {},
     }) as any
 
+  // altimate_change start — small calls must use thinking controls supported by each Gemini generation
+  for (const provider of [
+    { providerID: "google", npm: "@ai-sdk/google" },
+    { providerID: "google-vertex", npm: "@ai-sdk/google-vertex" },
+  ] as const) {
+    for (const testCase of [
+      { id: "gemini-2.0-flash", thinkingConfig: { thinkingBudget: 0 } },
+      { id: "gemini-2.5-flash", thinkingConfig: { thinkingBudget: 0 } },
+      { id: "gemini-3-pro", thinkingConfig: { thinkingLevel: "minimal" } },
+      { id: "gemini-3.1-pro", thinkingConfig: { thinkingLevel: "minimal" } },
+      { id: "models/gemini-3-pro-preview", thinkingConfig: { thinkingLevel: "minimal" } },
+      { id: "vertex_gemini-3pro", thinkingConfig: { thinkingLevel: "minimal" } },
+      { id: "vertex_gemini-4argon", thinkingConfig: { thinkingLevel: "low" } },
+      { id: "publishers/google/models/gemini-4-argon", thinkingConfig: { thinkingLevel: "low" } },
+      { id: "gemini-2.5-flash-lite-preview-06-17", thinkingConfig: { thinkingBudget: 0 } },
+      { id: "gemini-4-argon", thinkingConfig: { thinkingLevel: "low" } },
+      { id: "gemini-5-pro", thinkingConfig: { thinkingLevel: "low" } },
+      { id: "gemini-30-pro", thinkingConfig: { thinkingLevel: "low" } },
+      { id: "gemini-flash-latest", thinkingConfig: { thinkingBudget: 0 } },
+    ]) {
+      for (const configured of [false, true]) {
+        test(`${provider.providerID} ${testCase.id} uses supported small thinking controls with ${configured ? "generated" : "empty"} variants`, () => {
+          const model = createGoogleModel(testCase.id, provider)
+          if (configured) model.variants = ProviderTransform.variants(model)
+
+          const result = ProviderTransform.smallOptions(model)
+
+          expect(result).toEqual({ thinkingConfig: testCase.thinkingConfig })
+          if (testCase.thinkingConfig.thinkingLevel) {
+            expect(result.thinkingConfig).not.toHaveProperty("thinkingBudget")
+          }
+        })
+      }
+    }
+
+    test(`${provider.providerID} uses the lowest configured Gemini 4 thinking level regardless of variant names or order`, () => {
+      const model = createGoogleModel("gemini-4-argon", {
+        ...provider,
+        variants: {
+          detailed: { thinkingConfig: { includeThoughts: true, thinkingLevel: "high" } },
+          balanced: { thinkingConfig: { includeThoughts: true, thinkingLevel: "medium" } },
+        },
+      })
+
+      expect(ProviderTransform.smallOptions(model)).toEqual({ thinkingConfig: { thinkingLevel: "medium" } })
+    })
+  }
+  // altimate_change end
+
   test("disables Gemini 2.5 thinking budget for small calls even when high is the first variant", () => {
     const result = ProviderTransform.smallOptions(
       createGoogleModel("gemini-2.5-pro", {