diff --git a/packages/ai/CHANGELOG.md b/packages/ai/CHANGELOG.md
index f12e9de30..fcad3ef8f 100644
--- a/packages/ai/CHANGELOG.md
+++ b/packages/ai/CHANGELOG.md
@@ -14,6 +14,7 @@
 - Fixed OpenCode Go MiniMax M2.7 failing with "Model does not support this protocol" by using the Anthropic Messages API
 - Fixed automatic retry not triggering for gateway `Unable to forward upstream stream` errors ([#10920](https://github.com/earendil-works/pi/issues/10920))
 - Fixed OpenAI-compatible responses showing thinking after the answer when a proxy such as LiteLLM sends reasoning and content in the same stream chunk ([#10926](https://github.com/earendil-works/pi/issues/10926))
+- Fixed Gemini 4 Argon requests on Google Vertex failing with "Thinking budget is not supported", and its thinking level being ignored on the Gemini API, by sending `thinkingLevel` for Gemini 4 and later models ([#10890](https://github.com/earendil-works/pi/issues/10890))
 
 ## [1.2.0] - 2026-10-12
 
diff --git a/packages/ai/src/api/google-shared.ts b/packages/ai/src/api/google-shared.ts
index f546fbb00..aef32f316 100644
--- a/packages/ai/src/api/google-shared.ts
+++ b/packages/ai/src/api/google-shared.ts
@@ -75,6 +75,8 @@ export function usesGoogleThinkingLevel<T extends GoogleApiType>(model: Model<T>
 		// Match Gemini 3 Pro/Flash IDs with or without a minor version, such as
 		// gemini-3-flash-preview, gemini-3.1-pro-preview, and gemini-3.8-flash.
 		/gemini-3(?:\.\d+)?-(?:pro|flash)/.test(id) ||
+		// Match Gemini 4 and later, such as gemini-4-argon (#10890).
+		/gemini-(?:[4-9]|[1-9]\d)/.test(id) ||
 		id === "gemini-flash-latest" ||
 		id === "gemini-flash-lite-latest" ||
 		// Match both hosted Gemma 4 naming forms: gemma-4-* and gemma4-*.
diff --git a/packages/ai/test/google-thinking-level-map.test.ts b/packages/ai/test/google-thinking-level-map.test.ts
index 49380b2d2..ccc7c7803 100644
--- a/packages/ai/test/google-thinking-level-map.test.ts
+++ b/packages/ai/test/google-thinking-level-map.test.ts
@@ -166,6 +166,13 @@ describe("Google thinking level maps", () => {
 		expect(payload.config?.thinkingConfig).toEqual({ thinkingBudget: 0 });
 	});
 
+	// Regression test for https://github.com/earendil-works/pi/issues/10890
+	it.each(googleAdapters)("uses thinking levels for Gemini 4 models on $name", async ({ capture }) => {
+		const payload = await capture("gemini-4-argon", {}, "medium");
+
+		expect(payload.config?.thinkingConfig).toEqual({ includeThoughts: true, thinkingLevel: "MEDIUM" });
+	});
+
 	it.each(["xhigh", "max"] as const)("maps Google Generative AI %s to a supported level", async (reasoning) => {
 		const payload = await captureGooglePayload(
 			googleModel("gemini-3.7-flash", { xhigh: "high", max: "high" }),