fix(providers): backport ChatGPT Web reasoning transport limits

changelog.d/fixes/14194-chatgpt-web-transport-limits.md

@@ -0,0 +1 @@
+- Backport the upstream v6.1.7 conservative 500,000-character ChatGPT Web reasoning preflight limit for Medium/High and Pro Extra High, including multipart stages and final parts. Instant, Pro mode, and Luna keep their existing limits. This partial backport retains the v4.0.7 vendor pin; the full refresh and live acceptance remain pending. Refs #14194.

open-sse/vendor/codex-chatgpt-web/chatgpt-web-models.ts

@@ -19,7 +19,9 @@ export const CHATGPT_WEB_INSTANT_AUTO_COMPACT_TOKEN_LIMIT = 32_000;
 export const CHATGPT_WEB_MEDIUM_HIGH_CONTEXT_WINDOW = 90_000;
 export const CHATGPT_WEB_MEDIUM_HIGH_AUTO_COMPACT_TOKEN_LIMIT = 80_000;
 export const CHATGPT_WEB_INSTANT_COMPOSER_CHAR_LIMIT = 211_256;
-export const CHATGPT_WEB_MEDIUM_HIGH_COMPOSER_CHAR_LIMIT = 1_048_572;
+// Partial backport of the conservative reasoning composer ceiling from upstream v6.1.7,
+// commit f9ad4ae83a579287105ad822dd0c3e0029b04ef6. The overall vendor remains v4.0.7.
+export const CHATGPT_WEB_MEDIUM_HIGH_COMPOSER_CHAR_LIMIT = 500_000;
 /** Hidden ChatGPT product prompt and Codex Native schema reserve included in usage estimates. */
 export const CHATGPT_WEB_PLATFORM_RESERVE_TOKENS = 8_192;
 /** Pro-account usable browser windows and separately measured one-message boundaries. */
@@ -33,7 +35,8 @@ export const CHATGPT_WEB_PRO_STANDARD_CONTEXT_WINDOW =
 export const CHATGPT_WEB_PRO_MODEL_CONTEXT_WINDOW =
   CHATGPT_WEB_PRO_MODEL_MESSAGE_TOKEN_LIMIT + CHATGPT_WEB_PLATFORM_RESERVE_TOKENS + 1;
 export const CHATGPT_WEB_PRO_INSTANT_COMPOSER_CHAR_LIMIT = 545_000;
-export const CHATGPT_WEB_PRO_REASONING_COMPOSER_CHAR_LIMIT = 1_045_000;
+export const CHATGPT_WEB_PRO_REASONING_COMPOSER_CHAR_LIMIT =
+  CHATGPT_WEB_MEDIUM_HIGH_COMPOSER_CHAR_LIMIT;
 export const CHATGPT_WEB_PRO_MODEL_COMPOSER_CHAR_LIMIT = 1_635_000;
 /**
  * The underlying Luna model owns this context window. ChatGPT Free's much smaller browser request

tests/unit/chatgpt-web-codex-transport-limits-14194.test.ts

@@ -0,0 +1,120 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+
+import { requireChatGptWebCodexRoute } from "../../open-sse/executors/chatgpt-web-codex/models.ts";
+import {
+  assertChatGptWebInputWithinLimits,
+  assertChatGptWebMultipartInputWithinLimits,
+} from "../../open-sse/vendor/codex-chatgpt-web/adapters/chatgpt-web/browser-worker.ts";
+
+// Conservative transport policy from upstream v6.1.7, commit f9ad4ae83a579287105ad822dd0c3e0029b04ef6.
+// These literals describe the preflight contract, not a new measurement of the remote service.
+const LIMIT_ERROR = {
+  name: "ChatGptWebAdapterError",
+  status: 400,
+  errorType: "invalid_request_error",
+  code: "context_length_exceeded",
+  retryable: false,
+};
+
+function preflight(
+  id: string,
+  proAvailable: boolean,
+  chars: number,
+  inputTokens = 1_000,
+  messageTokens = 1_000
+) {
+  const route = requireChatGptWebCodexRoute(id);
+  assertChatGptWebInputWithinLimits(
+    inputTokens,
+    messageTokens,
+    route.backendModel,
+    route.effort,
+    { localToolsEnabled: true, solAvailable: route.sol, proAvailable },
+    chars
+  );
+}
+
+for (const [id, proAvailable] of [
+  ["medium", false],
+  ["high", false],
+  ["medium", true],
+  ["high", true],
+  ["extra-high", true],
+] as const) {
+  test(`${id} on ${proAvailable ? "Pro" : "Plus"} accepts 500k chars and rejects the next char`, () => {
+    const model = `chatgpt-web-codex/${id}`;
+    assert.doesNotThrow(() => preflight(model, proAvailable, 500_000));
+    assert.throws(() => preflight(model, proAvailable, 500_001), LIMIT_ERROR);
+  });
+}
+
+for (const [id, proAvailable, limit] of [
+  ["instant", false, 211_256],
+  ["instant", true, 545_000],
+  ["pro", true, 1_635_000],
+] as const) {
+  test(`${id} on ${proAvailable ? "Pro" : "Plus"} retains its distinct character boundary`, () => {
+    assert.doesNotThrow(() => preflight(id, proAvailable, limit));
+    assert.throws(() => preflight(id, proAvailable, limit + 1), LIMIT_ERROR);
+  });
+}
+
+for (const id of ["luna", "think"]) {
+  test(`${id} retains its token limit without acquiring a reasoning character limit`, () => {
+    assert.doesNotThrow(() => preflight(id, false, 500_001, 28_000));
+    assert.throws(() => preflight(id, false, 1, 28_001), LIMIT_ERROR);
+  });
+}
+
+test("the character ceiling does not replace Plus context-token rejection", () => {
+  assert.doesNotThrow(() => preflight("medium", false, 1, 89_999));
+  assert.throws(() => preflight("medium", false, 1, 90_000), LIMIT_ERROR);
+});
+
+test("the character ceiling does not replace Pro message-token rejection", () => {
+  assert.doesNotThrow(() => preflight("high", true, 1, 104_000, 103_000));
+  assert.throws(() => preflight("high", true, 1, 104_000, 103_001), LIMIT_ERROR);
+});
+
+test("unknown backend models remain rejected before transport", () => {
+  assert.throws(
+    () =>
+      assertChatGptWebInputWithinLimits(
+        1_000,
+        1_000,
+        "unknown-model",
+        "medium",
+        { localToolsEnabled: true, solAvailable: true, proAvailable: false },
+        1
+      ),
+    /context limit is not defined for model: unknown-model/
+  );
+});
+
+for (const proAvailable of [false, true]) {
+  for (const oversizedPart of ["stage", "final"] as const) {
+    test(`multipart ${oversizedPart} on ${proAvailable ? "Pro" : "Plus"} respects the 500k boundary`, () => {
+      const route = requireChatGptWebCodexRoute("high");
+      const checkMultipart = (chars: number) =>
+        assertChatGptWebMultipartInputWithinLimits(
+          1_000,
+          1_000,
+          route.backendModel,
+          route.effort,
+          { localToolsEnabled: true, solAvailable: true, proAvailable },
+          chars,
+          2,
+          {
+            stagingEffort: "medium",
+            maxStageMessageTokens: 1_000,
+            maxStageChars: oversizedPart === "stage" ? chars : 1,
+            finalMessageTokens: 1_000,
+            finalMessageChars: oversizedPart === "final" ? chars : 1,
+          }
+        );
+      assert.doesNotThrow(() => checkMultipart(500_000));
+      assert.throws(() => checkMultipart(500_001), LIMIT_ERROR);
+    });
+  }
+}