🤖 feat: update models to GPT-5.2 (#1103)

Updates the model definitions for the newly released GPT-5.2.

## Changes

- **GPT base model**: Updated from `gpt-5.1` to `gpt-5.2`
- **GPT-5.2 Pro**: Added new model with `medium`, `high`, `xhigh`
reasoning levels
- **Codex models**: Remain on 5.1 (`gpt-5.1-codex`,
`gpt-5.1-codex-mini`, `gpt-5.1-codex-max`)
- **Model data**: Added pricing/context info for `gpt-5.2` and
`gpt-5.2-pro` in models-extra.ts
- **Thinking policy**: Updated for gpt-5.2-pro's 3-level reasoning
support
- **Aliases**: `gpt` → `openai:gpt-5.2`, `gpt-pro` →
`openai:gpt-5.2-pro`

_Generated with `mux`_

Author: ammar-agent

docs/models.mdx

@@ -20,8 +20,8 @@ mux ships with a curated set of first-class models that we keep up to date with
 | Opus 4.5             | anthropic:claude-opus-4-5   | `opus`                                   | ✓       |
 | Sonnet 4.5           | anthropic:claude-sonnet-4-5 | `sonnet`                                 |         |
 | Haiku 4.5            | anthropic:claude-haiku-4-5  | `haiku`                                  |         |
-| GPT-5.1              | openai:gpt-5.1              | `gpt-5.1`                                |         |
-| GPT-5 Pro            | openai:gpt-5-pro            | `gpt-5-pro`                              |         |
+| GPT-5.2              | openai:gpt-5.2              | `gpt`                                    |         |
+| GPT-5.2 Pro          | openai:gpt-5.2-pro          | `gpt-pro`                                |         |
 | GPT-5.1 Codex        | openai:gpt-5.1-codex        | `codex`                                  |         |
 | GPT-5.1 Codex Mini   | openai:gpt-5.1-codex-mini   | `codex-mini`                             |         |
 | GPT-5.1 Codex Max    | openai:gpt-5.1-codex-max    | `codex-max`                              |         |

src/browser/utils/thinking/policy.test.ts

@@ -52,15 +52,44 @@ describe("getThinkingPolicyForModel", () => {
     ]);
   });
 
-  test("returns single HIGH for gpt-5-pro base model", () => {
+  test("returns medium/high/xhigh for gpt-5.2-pro", () => {
+    expect(getThinkingPolicyForModel("openai:gpt-5.2-pro")).toEqual(["medium", "high", "xhigh"]);
+  });
+
+  test("returns medium/high/xhigh for gpt-5.2-pro behind mux-gateway", () => {
+    expect(getThinkingPolicyForModel("mux-gateway:openai/gpt-5.2-pro")).toEqual([
+      "medium",
+      "high",
+      "xhigh",
+    ]);
+  });
+
+  test("returns 5 levels including xhigh for gpt-5.1-codex-max behind mux-gateway", () => {
+    expect(getThinkingPolicyForModel("mux-gateway:openai/gpt-5.1-codex-max")).toEqual([
+      "off",
+      "low",
+      "medium",
+      "high",
+      "xhigh",
+    ]);
+  });
+  test("returns medium/high/xhigh for gpt-5.2-pro with version suffix", () => {
+    expect(getThinkingPolicyForModel("openai:gpt-5.2-pro-2025-12-11")).toEqual([
+      "medium",
+      "high",
+      "xhigh",
+    ]);
+  });
+
+  test("returns single HIGH for gpt-5-pro base model (legacy)", () => {
     expect(getThinkingPolicyForModel("openai:gpt-5-pro")).toEqual(["high"]);
   });
 
-  test("returns single HIGH for gpt-5-pro with version suffix", () => {
+  test("returns single HIGH for gpt-5-pro with version suffix (legacy)", () => {
     expect(getThinkingPolicyForModel("openai:gpt-5-pro-2025-10-06")).toEqual(["high"]);
   });
 
-  test("returns single HIGH for gpt-5-pro with whitespace after colon", () => {
+  test("returns single HIGH for gpt-5-pro with whitespace after colon (legacy)", () => {
     expect(getThinkingPolicyForModel("openai: gpt-5-pro")).toEqual(["high"]);
   });
 

src/browser/utils/thinking/policy.ts

@@ -25,30 +25,43 @@ export type ThinkingPolicy = readonly ThinkingLevel[];
  *
  * Rules:
  * - openai:gpt-5.1-codex-max → ["off", "low", "medium", "high", "xhigh"] (5 levels including xhigh)
- * - openai:gpt-5-pro → ["high"] (only supported level)
+ * - openai:gpt-5.2-pro → ["medium", "high", "xhigh"] (3 levels)
+ * - openai:gpt-5-pro → ["high"] (only supported level, legacy)
  * - gemini-3 → ["low", "high"] (thinking level only)
  * - default → ["off", "low", "medium", "high"] (standard 4 levels)
  *
  * Tolerates version suffixes (e.g., gpt-5-pro-2025-10-06).
  * Does NOT match gpt-5-pro-mini (uses negative lookahead).
  */
 export function getThinkingPolicyForModel(modelString: string): ThinkingPolicy {
-  // Normalize to be robust to provider prefixes, whitespace, and version suffixes
+  // Normalize to be robust to provider prefixes, whitespace, gateway wrappers, and version suffixes
   const normalized = modelString.trim().toLowerCase();
   const withoutPrefix = normalized.replace(/^[a-z0-9_-]+:\s*/, "");
 
+  // Many providers/proxies encode the upstream provider as a path segment:
+  //   mux-gateway:openai/gpt-5.2-pro -> openai/gpt-5.2-pro -> gpt-5.2-pro
+  const withoutProviderNamespace = withoutPrefix.replace(/^[a-z0-9_-]+\//, "");
+
   // GPT-5.1-Codex-Max supports 5 reasoning levels including xhigh (Extra High)
-  if (withoutPrefix.startsWith("gpt-5.1-codex-max") || withoutPrefix.startsWith("codex-max")) {
+  if (
+    withoutProviderNamespace.startsWith("gpt-5.1-codex-max") ||
+    withoutProviderNamespace.startsWith("codex-max")
+  ) {
     return ["off", "low", "medium", "high", "xhigh"];
   }
 
-  // gpt-5-pro (not mini) with optional version suffix
-  if (/^gpt-5-pro(?!-[a-z])/.test(withoutPrefix)) {
+  // gpt-5.2-pro supports medium, high, xhigh reasoning levels
+  if (/^gpt-5\.2-pro(?!-[a-z])/.test(withoutProviderNamespace)) {
+    return ["medium", "high", "xhigh"];
+  }
+
+  // gpt-5-pro (legacy) only supports high
+  if (/^gpt-5-pro(?!-[a-z])/.test(withoutProviderNamespace)) {
     return ["high"];
   }
 
   // Gemini 3 Pro only supports "low" and "high" reasoning levels
-  if (withoutPrefix.includes("gemini-3")) {
+  if (withoutProviderNamespace.includes("gemini-3")) {
     return ["low", "high"];
   }
 

src/common/constants/knownModels.ts

@@ -48,15 +48,15 @@ const MODEL_DEFINITIONS = {
   },
   GPT: {
     provider: "openai",
-    providerModelId: "gpt-5.1",
-    aliases: ["gpt-5.1"],
+    providerModelId: "gpt-5.2",
+    aliases: ["gpt"],
     warm: true,
     tokenizerOverride: "openai/gpt-5",
   },
   GPT_PRO: {
     provider: "openai",
-    providerModelId: "gpt-5-pro",
-    aliases: ["gpt-5-pro"],
+    providerModelId: "gpt-5.2-pro",
+    aliases: ["gpt-pro"],
   },
   GPT_CODEX: {
     provider: "openai",

src/common/utils/ai/cacheStrategy.test.ts

@@ -26,7 +26,7 @@ describe("cacheStrategy", () => {
       expect(supportsAnthropicCache("openai:gpt-4")).toBe(false);
       expect(supportsAnthropicCache("google:gemini-2.0")).toBe(false);
       expect(supportsAnthropicCache("openrouter:meta-llama/llama-3.1")).toBe(false);
-      expect(supportsAnthropicCache("mux-gateway:openai/gpt-5.1")).toBe(false);
+      expect(supportsAnthropicCache("mux-gateway:openai/gpt-5.2")).toBe(false);
     });
   });
 

src/common/utils/tokens/displayUsage.test.ts

@@ -159,7 +159,7 @@ describe("createDisplayUsage", () => {
     };
 
     test("subtracts cached tokens for direct OpenAI model", () => {
-      const result = createDisplayUsage(openAIUsage, "openai:gpt-5.1");
+      const result = createDisplayUsage(openAIUsage, "openai:gpt-5.2");
 
       expect(result).toBeDefined();
       expect(result!.cached.tokens).toBe(71600);
@@ -169,7 +169,7 @@ describe("createDisplayUsage", () => {
 
     test("subtracts cached tokens for gateway OpenAI model", () => {
       // Gateway format: mux-gateway:openai/model-name
-      const result = createDisplayUsage(openAIUsage, "mux-gateway:openai/gpt-5.1");
+      const result = createDisplayUsage(openAIUsage, "mux-gateway:openai/gpt-5.2");
 
       expect(result).toBeDefined();
       expect(result!.cached.tokens).toBe(71600);
@@ -245,7 +245,7 @@ describe("createDisplayUsage", () => {
   });
 
   test("returns undefined for undefined usage", () => {
-    expect(createDisplayUsage(undefined, "openai:gpt-5.1")).toBeUndefined();
+    expect(createDisplayUsage(undefined, "openai:gpt-5.2")).toBeUndefined();
   });
 
   test("handles zero cached tokens", () => {
@@ -256,7 +256,7 @@ describe("createDisplayUsage", () => {
       cachedInputTokens: 0,
     };
 
-    const result = createDisplayUsage(usage, "openai:gpt-5.1");
+    const result = createDisplayUsage(usage, "openai:gpt-5.2");
 
     expect(result).toBeDefined();
     expect(result!.input.tokens).toBe(1000);
@@ -270,7 +270,7 @@ describe("createDisplayUsage", () => {
       totalTokens: 1500,
     };
 
-    const result = createDisplayUsage(usage, "openai:gpt-5.1");
+    const result = createDisplayUsage(usage, "openai:gpt-5.2");
 
     expect(result).toBeDefined();
     expect(result!.input.tokens).toBe(1000);

src/common/utils/tokens/displayUsage.ts

@@ -30,7 +30,7 @@ export function createDisplayUsage(
   const cachedTokens = usage.cachedInputTokens ?? 0;
   const rawInputTokens = usage.inputTokens ?? 0;
 
-  // Normalize gateway models (e.g., "mux-gateway:openai/gpt-5.1" → "openai:gpt-5.1")
+  // Normalize gateway models (e.g., "mux-gateway:openai/gpt-5.2" → "openai:gpt-5.2")
   // before detecting provider, so gateway-routed requests get correct handling
   const normalizedModel = normalizeGatewayModel(model);
 

src/common/utils/tokens/modelStats.test.ts

@@ -110,7 +110,7 @@ describe("getModelStats", () => {
 
   describe("model without provider prefix", () => {
     test("should handle model string without provider", () => {
-      const stats = getModelStats("gpt-5.1");
+      const stats = getModelStats("gpt-5.2");
       expect(stats).not.toBeNull();
       expect(stats?.max_input_tokens).toBeGreaterThan(0);
     });

src/common/utils/tokens/models-extra.ts

@@ -40,21 +40,40 @@ export const modelsExtra: Record<string, ModelData> = {
     supports_response_schema: true,
   },
 
-  // GPT-5 Pro - Released October 6, 2025 at DevDay
-  // $15/M input, $120/M output
-  // Only available via OpenAI's Responses API
-  "gpt-5-pro": {
+  // GPT-5.2 - Released December 11, 2025
+  // $1.75/M input, $14/M output
+  // Cached input: $0.175/M
+  "gpt-5.2": {
     max_input_tokens: 400000,
-    max_output_tokens: 272000,
-    input_cost_per_token: 0.000015, // $15 per million input tokens
-    output_cost_per_token: 0.00012, // $120 per million output tokens
+    max_output_tokens: 128000,
+    input_cost_per_token: 0.00000175, // $1.75 per million input tokens
+    output_cost_per_token: 0.000014, // $14 per million output tokens
+    // OpenAI model page lists "cached input" pricing, which corresponds to prompt cache reads.
+    cache_read_input_token_cost: 0.000000175, // $0.175 per million cached input tokens
+    litellm_provider: "openai",
+    mode: "chat",
+    supports_function_calling: true,
+    supports_vision: true,
+    supports_reasoning: true,
+    supports_response_schema: true,
+    knowledge_cutoff: "2025-08-31",
+  },
+
+  // GPT-5.2 Pro - Released December 11, 2025
+  // $21/M input, $168/M output
+  // Supports medium, high, xhigh reasoning levels
+  "gpt-5.2-pro": {
+    max_input_tokens: 400000,
+    max_output_tokens: 128000,
+    input_cost_per_token: 0.000021, // $21 per million input tokens
+    output_cost_per_token: 0.000168, // $168 per million output tokens
+    knowledge_cutoff: "2025-08-31",
     litellm_provider: "openai",
     mode: "chat",
     supports_function_calling: true,
     supports_vision: true,
     supports_reasoning: true,
     supports_response_schema: true,
-    knowledge_cutoff: "2024-09-30",
     supported_endpoints: ["/v1/responses"],
   },