From c2c32febb02c26076e7ef8c89aacfbf7a661d5c2 Mon Sep 17 00:00:00 2001
From: Armin Ronacher <armin.ronacher@active-4.com>
Date: Thu, 16 Jul 2026 18:56:54 +0200
Subject: [PATCH] fix(ai): correct Kimi K3 gateway output limits

---
 packages/ai/CHANGELOG.md                              |  4 ++++
 packages/ai/scripts/generate-models.ts                |  9 +++++++++
 packages/ai/src/providers/openrouter.models.ts        | 10 +++++-----
 packages/ai/src/providers/vercel-ai-gateway.models.ts |  2 +-
 4 files changed, 19 insertions(+), 6 deletions(-)

diff --git a/packages/ai/CHANGELOG.md b/packages/ai/CHANGELOG.md
index 5beb7f66..54c4d49f 100644
--- a/packages/ai/CHANGELOG.md
+++ b/packages/ai/CHANGELOG.md
@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Fixed Kimi K3 output limits for Vercel AI Gateway and OpenRouter models.
+
 ## [0.80.8] - 2026-07-16
 
 ### Breaking Changes
diff --git a/packages/ai/scripts/generate-models.ts b/packages/ai/scripts/generate-models.ts
index cd18e7f2..580ff04d 100644
--- a/packages/ai/scripts/generate-models.ts
+++ b/packages/ai/scripts/generate-models.ts
@@ -234,6 +234,8 @@ const KIMI_K3_THINKING_LEVEL_MAP = {
 	xhigh: null,
 	max: "max",
 } as const;
+const KIMI_K3_MAX_TOKENS = 131072;
+const OPENROUTER_KIMI_K3_MODEL_IDS = new Set(["moonshotai/kimi-k3", "~moonshotai/kimi-latest"]);
 
 const ANT_LING_RING_THINKING_LEVEL_MAP = {
 	off: null,
@@ -1825,6 +1827,13 @@ async function generateModels() {
 		if (candidate.provider === "openai" && candidate.id === "gpt-5-pro") {
 			candidate.maxTokens = 128000;
 		}
+		// Keep Kimi K3's canonical output limit when gateway metadata is missing or incorrect.
+		if (
+			(candidate.provider === "openrouter" && OPENROUTER_KIMI_K3_MODEL_IDS.has(candidate.id)) ||
+			(candidate.provider === "vercel-ai-gateway" && candidate.id === "moonshotai/kimi-k3")
+		) {
+			candidate.maxTokens = KIMI_K3_MAX_TOKENS;
+		}
 		// Keep selected OpenRouter model metadata stable until upstream settles.
 		if (candidate.provider === "openrouter" && candidate.id === "moonshotai/kimi-k2.5") {
 			candidate.cost.input = 0.41;
diff --git a/packages/ai/src/providers/openrouter.models.ts b/packages/ai/src/providers/openrouter.models.ts
index 142c447b..03f2ada1 100644
--- a/packages/ai/src/providers/openrouter.models.ts
+++ b/packages/ai/src/providers/openrouter.models.ts
@@ -1939,7 +1939,7 @@ export const OPENROUTER_MODELS = {
 			cacheWrite: 0,
 		},
 		contextWindow: 1048576,
-		maxTokens: 4096,
+		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
 	"nex-agi/nex-n2-mini": {
 		id: "nex-agi/nex-n2-mini",
@@ -4706,9 +4706,9 @@ export const OPENROUTER_MODELS = {
 		thinkingLevelMap: {"xhigh":"xhigh"},
 		input: ["text"],
 		cost: {
-			input: 0.9338,
-			output: 2.9348,
-			cacheRead: 0.17342,
+			input: 0.9226,
+			output: 2.8996,
+			cacheRead: 0.17134,
 			cacheWrite: 0,
 		},
 		contextWindow: 1048576,
@@ -4856,7 +4856,7 @@ export const OPENROUTER_MODELS = {
 			cacheWrite: 0,
 		},
 		contextWindow: 1048576,
-		maxTokens: 4096,
+		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
 	"~openai/gpt-latest": {
 		id: "~openai/gpt-latest",
diff --git a/packages/ai/src/providers/vercel-ai-gateway.models.ts b/packages/ai/src/providers/vercel-ai-gateway.models.ts
index e5fc7f1c..dccf0695 100644
--- a/packages/ai/src/providers/vercel-ai-gateway.models.ts
+++ b/packages/ai/src/providers/vercel-ai-gateway.models.ts
@@ -1905,7 +1905,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
-		maxTokens: 256000,
+		maxTokens: 131072,
 	} satisfies Model<"anthropic-messages">,
 	"nvidia/nemotron-3-nano-30b-a3b": {
 		id: "nvidia/nemotron-3-nano-30b-a3b",