From bcbad6f8b10047090d3c85f18bebcfa557ced122 Mon Sep 17 00:00:00 2001
From: Cristina Poncela Cubeiro
 <140309543+cristinaponcela@users.noreply.github.com>
Date: Wed, 28 Oct 2026 15:45:03 +0100
Subject: [PATCH] fix(ai): detect Bedrock Mantle context overflow errors

---
 packages/ai/src/utils/overflow.ts | 3 +++
 packages/ai/test/overflow.test.ts | 7 +++++++
 2 files changed, 10 insertions(+)

diff --git a/packages/ai/src/utils/overflow.ts b/packages/ai/src/utils/overflow.ts
index 36ae26b49..8caaeb8e4 100644
--- a/packages/ai/src/utils/overflow.ts
+++ b/packages/ai/src/utils/overflow.ts
@@ -14,6 +14,7 @@ import type { AssistantMessage } from "../types.ts";
  * - OpenAI/LiteLLM: "Requested token count exceeds the model's maximum context length of 131072 tokens"
  * - LiteLLM pre-call check: "litellm.ContextWindowExceededError: ... Context Window exceeded for given call."
  * - OpenAI-compatible: "Input length (265330) exceeds model's maximum context length (262144)."
+ * - Amazon Bedrock Mantle (OpenAI Responses): "invalid_prompt: prompt tokens (330095) exceed model maximum (278528) for base model ..."
  * - Google: "The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)"
  * - xAI: "This model's maximum prompt length is 131072 but the request contains 537812 tokens"
  * - Groq: "Please reduce the length of the messages or completion"
@@ -43,6 +44,7 @@ const OVERFLOW_PATTERNS = [
 	/input is too long for requested model/i, // Amazon Bedrock
 	/exceeds the context window/i, // OpenAI (Completions & Responses API)
 	/exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\d,]+ tokens?|\s*\([\d,]+\))/i, // OpenAI-compatible proxies (LiteLLM)
+	/prompt tokens \([\d,]+\) exceed model maximum/i, // Amazon Bedrock Mantle (OpenAI Responses observed error; exact text is not documented by AWS)
 	/context window exceeded/i, // LiteLLM pre-call check (ContextWindowExceededError)
 	/input token count.*exceeds the maximum/i, // Google (Gemini)
 	/maximum prompt length is \d+/i, // xAI (Grok)
@@ -101,6 +103,7 @@ const NON_OVERFLOW_PATTERNS = [
  * **Reliable detection (returns error with detectable message):**
  * - Anthropic: "prompt is too long: X tokens > Y maximum" or "request_too_large"
  * - OpenAI (Completions & Responses): "exceeds the context window", "exceeds the model's maximum context length of X tokens", or "exceeds model's maximum context length (X)"
+ * - Amazon Bedrock Mantle (OpenAI Responses): "invalid_prompt: prompt tokens (X) exceed model maximum (Y) for base model ..."
  * - Google Gemini: "input token count exceeds the maximum"
  * - xAI (Grok): "maximum prompt length is X but request contains Y"
  * - Groq: "reduce the length of the messages"
diff --git a/packages/ai/test/overflow.test.ts b/packages/ai/test/overflow.test.ts
index 1cab5e92e..bd2a454ea 100644
--- a/packages/ai/test/overflow.test.ts
+++ b/packages/ai/test/overflow.test.ts
@@ -106,6 +106,13 @@ describe("isContextOverflow", () => {
 		expect(isContextOverflow(message, 262144)).toBe(true);
 	});
 
+	it("detects Amazon Bedrock Mantle OpenAI Responses prompt-token maximum errors", () => {
+		const message = createErrorMessage(
+			"Amazon Bedrock Mantle OpenAI Responses API error (400): invalid_prompt: prompt tokens (330095) exceed model maximum (278528) for base model abc",
+		);
+		expect(isContextOverflow(message, 272000)).toBe(true);
+	});
+
 	it("detects OpenRouter Poolside maximum allowed input length errors", () => {
 		const message = createErrorMessage(
 			"Provider returned error: Input length 131393 exceeds the maximum allowed input length of 131040 tokens.",