Skip to content

Never stop coding. Free MIT AI gateway: one endpoint, 375 providers (150+ free), 1,400+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 900+ contributors

fix(routing): honor client reasoning.effort for gpt-5.5 + route suffixed variants to codex (#2877)

diegosouzapwf5d74cd76dfd
diff --git a/CHANGELOG.md b/CHANGELOG.md
index e52c402baf..cf4a57adb4 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -59,6 +59,14 @@
### Fixed
+- **routing/codex:** fix two gpt-5.5 Codex defects (#2877). (A) For a Codex-only
+ account, a bare `gpt-5.5` Responses request was rerouted to codex with the
+ model hardcoded to `gpt-5.5-medium` (`chatHelpers.ts`); the executor read that
+ `-medium` suffix as an explicit `modelEffort` that (per #2331) overrode a
+ client `reasoning.effort=xhigh`, silently demoting it — now it keeps the bare
+ `gpt-5.5` id so the client effort wins. (B) `gpt-5.5-xhigh`/`-high`/`-low`
+ misrouted to `openai` (→ "No credentials" for codex-only users); the suffixed
+ variants are now in `CODEX_PREFERRED_UNPREFIXED_MODELS` so they infer codex.
- **sse/chatCore:** remove a duplicate `const settings` declaration in
`handleChatCore` (introduced alongside the per-key stream-default-mode
feature). The same-scope redeclaration made esbuild/tsx fail with
diff --git a/open-sse/services/model.ts b/open-sse/services/model.ts
index 5c15c0ac84..2f3ca61d1a 100644
--- a/open-sse/services/model.ts
+++ b/open-sse/services/model.ts
@@ -112,7 +112,16 @@ for (const [aliasOrId, models] of Object.entries(PROVIDER_MODELS)) {
}
}
const KNOWN_MODEL_IDS = new Set(MODEL_TO_PROVIDERS.keys());
-const CODEX_PREFERRED_UNPREFIXED_MODELS = new Set(["gpt-5.5"]);
+// #2877(B): include the effort-suffixed variants so a bare `gpt-5.5-xhigh`
+// (and -high/-medium/-low) infers the codex provider instead of falling through
+// the `/^gpt-/` → openai fallback (which 500s for codex-only credentials).
+const CODEX_PREFERRED_UNPREFIXED_MODELS = new Set([
+ "gpt-5.5",
+ "gpt-5.5-xhigh",
+ "gpt-5.5-high",
+ "gpt-5.5-medium",
+ "gpt-5.5-low",
+]);
const CODEX_PREFERRED_UNPREFIXED_MODEL_ALIASES = new Map([["gpt-5.5", "gpt-5.5-medium"]]);
export const CODEX_NATIVE_UNPREFIXED_MODELS = new Set(["codex-auto-review"]);
diff --git a/src/sse/handlers/chatHelpers.ts b/src/sse/handlers/chatHelpers.ts
index 15e4bb2d7f..ebaa542927 100644
--- a/src/sse/handlers/chatHelpers.ts
+++ b/src/sse/handlers/chatHelpers.ts
@@ -130,9 +130,14 @@ export async function resolveModelOrError(
!isCodexNativeResponsesRequest(body, endpointPath, requestHeaders) &&
(await hasOnlyActiveCodexAccount())
) {
- log.info("ROUTING", `${modelStr} → codex/gpt-5.5-medium (Codex-only active account)`);
+ // #2877: keep the bare model id (do NOT bake a `-medium` suffix). The Codex
+ // executor reads a model-name suffix as an explicit `modelEffort` that (per
+ // #2331) overrides the client's `reasoning.effort`, so injecting `-medium`
+ // here silently demoted a genuine `reasoning.effort=xhigh`. The default
+ // effort still comes from the connection fallback when the client sends none.
+ log.info("ROUTING", `${modelStr} → codex/gpt-5.5 (Codex-only active account)`);
modelInfo.provider = "codex";
- modelInfo.model = "gpt-5.5-medium";
+ modelInfo.model = "gpt-5.5";
}
// Forced-rewrite: codex provider doesn't serve DeepSeek/Qwen/Kimi/etc. Reroute
diff --git a/tests/unit/codex-gpt55-effort-routing.test.ts b/tests/unit/codex-gpt55-effort-routing.test.ts
new file mode 100644
index 0000000000..17b0abdb40
--- /dev/null
+++ b/tests/unit/codex-gpt55-effort-routing.test.ts
@@ -0,0 +1,71 @@
+/**
+ * Issue #2877 — two routing/effort defects for the gpt-5.5 Codex family:
+ *
+ * (B) `gpt-5.5-xhigh` (and -high/-low) misrouted to the `openai` provider
+ * because only bare `gpt-5.5` was in CODEX_PREFERRED_UNPREFIXED_MODELS — the
+ * suffixed variants fell through to the `/^gpt-/` → openai fallback, so a
+ * Codex-OAuth-only user got "No credentials for provider: openai". Fixed by
+ * adding the variants to the set (`open-sse/services/model.ts`).
+ *
+ * (A) For a Codex-only account, a bare `gpt-5.5` Responses request was rerouted
+ * to codex but with the model hardcoded to `gpt-5.5-medium`
+ * (`src/sse/handlers/chatHelpers.ts`). The Codex executor reads that `-medium`
+ * suffix as an explicit `modelEffort`, which (per #2331) overrides the
+ * client's `reasoning.effort=xhigh` — silently demoting it. Fixed by keeping
+ * the bare `gpt-5.5` id; the executor's modelEffort-first precedence (#2331)
+ * is left untouched.
+ */
+import test from "node:test";
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-gpt55-routing-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+
+const core = await import("../../src/lib/db/core.ts");
+const providersDb = await import("../../src/lib/db/providers.ts");
+const { getModelInfoCore } = await import("../../open-sse/services/model.ts");
+const { resolveModelOrError } = await import("../../src/sse/handlers/chatHelpers.ts");
+
+test.before(async () => {
+ // Codex-only active account (no openai connection).
+ await providersDb.createProviderConnection({
+ provider: "codex",
+ authType: "oauth",
+ email: "[email protected]",
+ providerSpecificData: { workspaceId: "ws-1" },
+ });
+});
+
+test.after(() => {
+ core.resetDbInstance();
+ fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});
+
+// ── Defect B: suffixed bare names infer codex, not openai ─────────────────────
+for (const variant of ["gpt-5.5-xhigh", "gpt-5.5-high", "gpt-5.5-low"]) {
+ test(`#2877(B) ${variant} infers codex (not openai)`, async () => {
+ const info = await getModelInfoCore(variant, null);
+ assert.equal(info.provider, "codex", `${variant} must infer the codex provider`);
+ assert.equal(info.model, variant, "the explicit effort suffix must be preserved");
+ });
+}
+
+// ── Defect A: Codex-only bare gpt-5.5 reroute must NOT bake a -medium suffix ───
+test("#2877(A) Codex-only bare gpt-5.5 Responses request keeps the bare model id", async () => {
+ const result = (await resolveModelOrError(
+ "gpt-5.5",
+ { input: [{ role: "user", content: [{ type: "input_text", text: "hi" }] }] },
+ "/v1/responses",
+ null
+ )) as { provider?: string; model?: string };
+
+ assert.equal(result.provider, "codex", "Codex-only account must reroute gpt-5.5 to codex");
+ assert.equal(
+ result.model,
+ "gpt-5.5",
+ "must NOT inject a -medium suffix (that would override a client reasoning.effort=xhigh)"
+ );
+});