fix(ai): restore Codex context limits

2026-06-13 11:02:38 +02:00
parent 032c01c1e1
commit aa3a5233ad
4 changed files with 86 additions and 98 deletions
--- a/packages/ai/CHANGELOG.md
+++ b/packages/ai/CHANGELOG.md
@@ -2,6 +2,10 @@

 ## [Unreleased]

+### Fixed
+
+- Restored OpenAI GPT-5.4/GPT-5.5 and OpenAI Codex GPT-5.4/GPT-5.4 mini/GPT-5.5 context window metadata to the observed 272k-token Codex backend limit, avoiding a billing hazard from sending prompts above Codex's accepted limit (reported by [@trethore](https://github.com/trethore)).
+
 ## [0.79.2] - 2026-06-12

 ### Added
--- a/packages/ai/scripts/generate-models.ts
+++ b/packages/ai/scripts/generate-models.ts
@@ -1367,7 +1367,7 @@ async function generateModels() {
 			candidate.maxTokens = 128000;
 		}
 		if (candidate.provider === "openai" && (candidate.id === "gpt-5.4" || candidate.id === "gpt-5.5")) {
-			candidate.contextWindow = 1050000;
+			candidate.contextWindow = 272000;
 			candidate.maxTokens = 128000;
 		}
 		// models.dev reports gpt-5-pro output as 272000 (a duplicate of the input sub-limit);
@@ -1634,7 +1634,7 @@ async function generateModels() {
 				cacheRead: 0.25,
 				cacheWrite: 0,
 			},
-			contextWindow: 1050000,
+			contextWindow: 272000,
 			maxTokens: 128000,
 		});
 	}
@@ -1761,9 +1761,9 @@ async function generateModels() {

 	// OpenAI Codex (ChatGPT OAuth) models
 	// NOTE: These are not fetched from models.dev; we keep a small, explicit list to avoid aliases.
+	// Context window is based on observed server limits (400s above ~272k), not marketing numbers.
 	const CODEX_BASE_URL = "https://chatgpt.com/backend-api";
-	const CODEX_GPT_54_CONTEXT = 1000000;
-	const CODEX_STANDARD_CONTEXT = 400000;
+	const CODEX_CONTEXT = 272000;
 	const CODEX_SPARK_CONTEXT = 128000;
 	const CODEX_MAX_TOKENS = 128000;
 	const codexModels: Model<"openai-codex-responses">[] = [
@@ -1788,7 +1788,7 @@ async function generateModels() {
 			reasoning: true,
 			input: ["text", "image"],
 			cost: { input: 2.5, output: 15, cacheRead: 0.25, cacheWrite: 0 },
-			contextWindow: CODEX_GPT_54_CONTEXT,
+			contextWindow: CODEX_CONTEXT,
 			maxTokens: CODEX_MAX_TOKENS,
 		},
 		{
@@ -1800,7 +1800,7 @@ async function generateModels() {
 			reasoning: true,
 			input: ["text", "image"],
 			cost: { input: 0.75, output: 4.5, cacheRead: 0.075, cacheWrite: 0 },
-			contextWindow: CODEX_STANDARD_CONTEXT,
+			contextWindow: CODEX_CONTEXT,
 			maxTokens: CODEX_MAX_TOKENS,
 		},
 		{
@@ -1812,7 +1812,7 @@ async function generateModels() {
 			reasoning: true,
 			input: ["text", "image"],
 			cost: { input: 5, output: 30, cacheRead: 0.5, cacheWrite: 0 },
-			contextWindow: CODEX_STANDARD_CONTEXT,
+			contextWindow: CODEX_CONTEXT,
 			maxTokens: CODEX_MAX_TOKENS,
 		},
 	];
--- a/packages/ai/src/models.generated.ts
+++ b/packages/ai/src/models.generated.ts
@@ -3945,24 +3945,6 @@ export const MODELS = {
 			contextWindow: 131072,
 			maxTokens: 32768,
 		} satisfies Model<"anthropic-messages">,
-		"accounts/fireworks/models/kimi-k2p5": {
-			id: "accounts/fireworks/models/kimi-k2p5",
-			name: "Kimi K2.5",
-			api: "anthropic-messages",
-			provider: "fireworks",
-			baseUrl: "https://api.fireworks.ai/inference",
-			compat: {"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},
-			reasoning: true,
-			input: ["text", "image"],
-			cost: {
-				input: 0.6,
-				output: 3,
-				cacheRead: 0.1,
-				cacheWrite: 0,
-			},
-			contextWindow: 256000,
-			maxTokens: 256000,
-		} satisfies Model<"anthropic-messages">,
 		"accounts/fireworks/models/kimi-k2p6": {
 			id: "accounts/fireworks/models/kimi-k2p6",
 			name: "Kimi K2.6",
@@ -3981,23 +3963,23 @@ export const MODELS = {
 			contextWindow: 262000,
 			maxTokens: 262000,
 		} satisfies Model<"anthropic-messages">,
-		"accounts/fireworks/models/minimax-m2p5": {
-			id: "accounts/fireworks/models/minimax-m2p5",
-			name: "MiniMax-M2.5",
+		"accounts/fireworks/models/kimi-k2p7-code": {
+			id: "accounts/fireworks/models/kimi-k2p7-code",
+			name: "Kimi K2.7 Code",
 			api: "anthropic-messages",
 			provider: "fireworks",
 			baseUrl: "https://api.fireworks.ai/inference",
 			compat: {"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},
 			reasoning: true,
-			input: ["text"],
+			input: ["text", "image"],
 			cost: {
-				input: 0.3,
-				output: 1.2,
-				cacheRead: 0.03,
+				input: 0.95,
+				output: 4,
+				cacheRead: 0.19,
 				cacheWrite: 0,
 			},
-			contextWindow: 196608,
-			maxTokens: 196608,
+			contextWindow: 262000,
+			maxTokens: 262000,
 		} satisfies Model<"anthropic-messages">,
 		"accounts/fireworks/models/minimax-m2p7": {
 			id: "accounts/fireworks/models/minimax-m2p7",
@@ -4017,9 +3999,27 @@ export const MODELS = {
 			contextWindow: 196608,
 			maxTokens: 196608,
 		} satisfies Model<"anthropic-messages">,
-		"accounts/fireworks/models/qwen3p6-plus": {
-			id: "accounts/fireworks/models/qwen3p6-plus",
-			name: "Qwen 3.6 Plus",
+		"accounts/fireworks/models/minimax-m3": {
+			id: "accounts/fireworks/models/minimax-m3",
+			name: "MiniMax-M3",
+			api: "anthropic-messages",
+			provider: "fireworks",
+			baseUrl: "https://api.fireworks.ai/inference",
+			compat: {"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0.3,
+				output: 1.2,
+				cacheRead: 0.06,
+				cacheWrite: 0,
+			},
+			contextWindow: 512000,
+			maxTokens: 512000,
+		} satisfies Model<"anthropic-messages">,
+		"accounts/fireworks/models/qwen3p7-plus": {
+			id: "accounts/fireworks/models/qwen3p7-plus",
+			name: "Qwen 3.7 Plus",
 			api: "anthropic-messages",
 			provider: "fireworks",
 			baseUrl: "https://api.fireworks.ai/inference",
@@ -4027,9 +4027,9 @@ export const MODELS = {
 			reasoning: true,
 			input: ["text", "image"],
 			cost: {
-				input: 0.5,
-				output: 3,
-				cacheRead: 0.1,
+				input: 0.4,
+				output: 1.6,
+				cacheRead: 0.08,
 				cacheWrite: 0,
 			},
 			contextWindow: 262144,
@@ -4089,6 +4089,24 @@ export const MODELS = {
 			contextWindow: 262000,
 			maxTokens: 262000,
 		} satisfies Model<"anthropic-messages">,
+		"accounts/fireworks/routers/kimi-k2p7-code-fast": {
+			id: "accounts/fireworks/routers/kimi-k2p7-code-fast",
+			name: "Kimi K2.7 Code Fast",
+			api: "anthropic-messages",
+			provider: "fireworks",
+			baseUrl: "https://api.fireworks.ai/inference",
+			compat: {"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},
+			reasoning: true,
+			input: ["text", "image"],
+			cost: {
+				input: 2,
+				output: 8,
+				cacheRead: 0.38,
+				cacheWrite: 0,
+			},
+			contextWindow: 262000,
+			maxTokens: 262000,
+		} satisfies Model<"anthropic-messages">,
 	},
 	"github-copilot": {
 		"claude-fable-5": {
@@ -7382,7 +7400,7 @@ export const MODELS = {
 				cacheRead: 0.25,
 				cacheWrite: 0,
 			},
-			contextWindow: 1050000,
+			contextWindow: 272000,
 			maxTokens: 128000,
 		} satisfies Model<"openai-responses">,
 		"gpt-5.4-mini": {
@@ -7454,7 +7472,7 @@ export const MODELS = {
 				cacheRead: 0.5,
 				cacheWrite: 0,
 			},
-			contextWindow: 1050000,
+			contextWindow: 272000,
 			maxTokens: 128000,
 		} satisfies Model<"openai-responses">,
 		"gpt-5.5-pro": {
@@ -7646,7 +7664,7 @@ export const MODELS = {
 				cacheRead: 0.25,
 				cacheWrite: 0,
 			},
-			contextWindow: 1000000,
+			contextWindow: 272000,
 			maxTokens: 128000,
 		} satisfies Model<"openai-codex-responses">,
 		"gpt-5.4-mini": {
@@ -7664,7 +7682,7 @@ export const MODELS = {
 				cacheRead: 0.075,
 				cacheWrite: 0,
 			},
-			contextWindow: 400000,
+			contextWindow: 272000,
 			maxTokens: 128000,
 		} satisfies Model<"openai-codex-responses">,
 		"gpt-5.5": {
@@ -7682,7 +7700,7 @@ export const MODELS = {
 				cacheRead: 0.5,
 				cacheWrite: 0,
 			},
-			contextWindow: 400000,
+			contextWindow: 272000,
 			maxTokens: 128000,
 		} satisfies Model<"openai-codex-responses">,
 	},
@@ -7705,25 +7723,6 @@ export const MODELS = {
 			contextWindow: 200000,
 			maxTokens: 32000,
 		} satisfies Model<"openai-completions">,
-		"claude-fable-5": {
-			id: "claude-fable-5",
-			name: "Claude Fable 5",
-			api: "anthropic-messages",
-			provider: "opencode",
-			baseUrl: "https://opencode.ai/zen",
-			compat: {"forceAdaptiveThinking":true},
-			reasoning: true,
-			thinkingLevelMap: {"off":null,"xhigh":"xhigh"},
-			input: ["text", "image"],
-			cost: {
-				input: 10,
-				output: 50,
-				cacheRead: 1,
-				cacheWrite: 12.5,
-			},
-			contextWindow: 1000000,
-			maxTokens: 128000,
-		} satisfies Model<"anthropic-messages">,
 		"claude-haiku-4-5": {
 			id: "claude-haiku-4-5",
 			name: "Claude Haiku 4.5",
@@ -13082,6 +13081,25 @@ export const MODELS = {
 			contextWindow: 202752,
 			maxTokens: 131072,
 		} satisfies Model<"openai-completions">,
+		"MiniMaxAI/MiniMax-M3": {
+			id: "MiniMaxAI/MiniMax-M3",
+			name: "MiniMax-M3",
+			api: "openai-completions",
+			provider: "together",
+			baseUrl: "https://api.together.ai/v1",
+			compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
+			reasoning: true,
+			thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
+			input: ["text", "image"],
+			cost: {
+				input: 0.3,
+				output: 1.2,
+				cacheRead: 0.06,
+				cacheWrite: 0,
+			},
+			contextWindow: 524288,
+			maxTokens: 250000,
+		} satisfies Model<"openai-completions">,
 		"Qwen/Qwen2.5-7B-Instruct-Turbo": {
 			id: "Qwen/Qwen2.5-7B-Instruct-Turbo",
 			name: "Qwen 2.5 7B Instruct Turbo",
@@ -13791,25 +13809,6 @@ export const MODELS = {
 			contextWindow: 200000,
 			maxTokens: 8192,
 		} satisfies Model<"anthropic-messages">,
-		"anthropic/claude-fable-5": {
-			id: "anthropic/claude-fable-5",
-			name: "Claude Fable 5",
-			api: "anthropic-messages",
-			provider: "vercel-ai-gateway",
-			baseUrl: "https://ai-gateway.vercel.sh",
-			compat: {"forceAdaptiveThinking":true},
-			reasoning: true,
-			thinkingLevelMap: {"off":null,"xhigh":"xhigh"},
-			input: ["text", "image"],
-			cost: {
-				input: 10,
-				output: 50,
-				cacheRead: 1,
-				cacheWrite: 12.5,
-			},
-			contextWindow: 1000000,
-			maxTokens: 128000,
-		} satisfies Model<"anthropic-messages">,
 		"anthropic/claude-haiku-4.5": {
 			id: "anthropic/claude-haiku-4.5",
 			name: "Claude Haiku 4.5",
--- a/packages/ai/test/openai-model-metadata.test.ts
+++ b/packages/ai/test/openai-model-metadata.test.ts
@@ -1,15 +0,0 @@
-import { describe, expect, it } from "vitest";
-import { getModel } from "../src/models.ts";
-
-describe("OpenAI model metadata", () => {
-	it("uses current GPT-5.4 and GPT-5.5 API context windows", () => {
-		expect(getModel("openai", "gpt-5.4").contextWindow).toBe(1050000);
-		expect(getModel("openai", "gpt-5.5").contextWindow).toBe(1050000);
-	});
-
-	it("uses current OpenAI Codex context windows", () => {
-		expect(getModel("openai-codex", "gpt-5.4").contextWindow).toBe(1000000);
-		expect(getModel("openai-codex", "gpt-5.4-mini").contextWindow).toBe(400000);
-		expect(getModel("openai-codex", "gpt-5.5").contextWindow).toBe(400000);
-	});
-});