codex-multi-auth 2.14.0 → 2.16.0

package/dist/lib/request/helpers/model-map.js

@@ -52,7 +52,21 @@ const TOOL_CAPABILITIES = {
         compaction: false,
     },
 };
-export const CURRENT_CODEX_MODEL = "gpt-5.3-codex";
+/**
+ * Where every codex-named id routes. No codex model is left: upstream Codex
+ * dropped `gpt-5.3-codex` from its catalog on 2026-07-08, and OpenAI's
+ * deprecations page names `gpt-5.6-sol` as the replacement for every retired
+ * codex id (`gpt-5-codex`, `gpt-5.1-codex*`, `gpt-5.2-codex`). The export
+ * keeps its name so callers that mean "the model a codex request runs on"
+ * still read naturally.
+ */
+export const CURRENT_CODEX_MODEL = "gpt-5.6-sol";
+/**
+ * The retired codex mini ids (`gpt-5.1-codex-mini`, `gpt-5-codex-mini`,
+ * `codex-mini-latest`) go to Terra, OpenAI's named replacement for
+ * `gpt-5.1-codex-mini`, rather than up to Sol.
+ */
+const CODEX_MINI_REPLACEMENT_MODEL = "gpt-5.6-terra";
 export const DEFAULT_MODEL = "gpt-5.5";
 // Model used for diagnostic live/quota probes (`check`, `report`, `best`).
 // Deliberately distinct from DEFAULT_MODEL: GPT-5.6 is the latest general family
@@ -70,18 +84,18 @@ export const DEFAULT_MODEL = "gpt-5.5";
 export const DEFAULT_PROBE_MODEL = "gpt-5.6-sol";
 // Single source of truth for the live/quota probe fallback chain. Both the
 // manager probe (lib/quota-probe.ts) and the runtime probe (lib/runtime/quota-probe.ts)
-// import this so the ordered candidate list cannot drift between them. It leads
-// with GPT-5.6 and steps down so accounts without 5.6 entitlement still resolve
-// a working probe model.
+// import this so the ordered candidate list cannot drift between them.
+//
+// Every entry must be a model the Codex backend still serves. The chain used to
+// end on `gpt-5.4` and three codex models; all four are gone from the upstream
+// catalog (`gpt-5.4` carries `retirement_at: 2026-08-31`), so a probe that
+// reached them spent requests on guaranteed failures. `gpt-6-luna` closes it
+// because it is offered on the most plans of any catalog model (24).
 export const QUOTA_PROBE_MODEL_CHAIN = [
     DEFAULT_PROBE_MODEL,
     DEFAULT_MODEL,
-    "gpt-5.4",
-    "gpt-5.3-codex",
-    "gpt-5.2-codex",
-    "gpt-5-codex",
+    "gpt-6-luna",
 ];
-const LEGACY_CODEX_MODEL = "gpt-5-codex";
 /**
  * GPT-5.6 tiers, per the upstream Codex catalog
  * (openai/codex `codex-rs/models-manager/models.json`).
@@ -112,8 +126,8 @@ const GPT_5_6_LUNA_EFFORTS = [
 /**
  * GPT-6 Astra, OpenAI's 2026-09-03 frontier release.
  *
- * There is no Sol/Terra/Luna split this generation: the announced lineup is the
- * flagship plus `aeon`, a long-horizon variant built for runs measured in days.
+ * The launch lineup was the flagship plus `aeon`, a long-horizon variant built
+ * for runs measured in days; Sol and Luna followed on 2026-09-22 (below).
  * `gpt-6-astra` is the API model name OpenAI published at launch;
  * `gpt-6-astra-aeon` is the second slug that shipped beside it in the Codex
  * model list. "Astra Pro" is a plan tier, not a separate slug we have seen, so
@@ -139,6 +153,24 @@ const GPT_6_ASTRA_EFFORTS = [
     "max",
     "ultra",
 ];
+/**
+ * GPT-6 Sol and Luna, added to the upstream Codex catalog on 2026-09-22
+ * (openai/codex #47332) beside Astra. Sol is the everyday workhorse and Luna
+ * the small, cheap tier; there is no GPT-6 Terra. Upstream migrates `gpt-5.5`,
+ * `gpt-5.6-sol` and `gpt-5.6-terra` users to Sol and `gpt-5.6-luna` users to
+ * Luna. Ladders and defaults below are the catalog's own values: Sol reaches
+ * `ultra`, Luna stops at `max` (same split as 5.6), both default to `medium`.
+ */
+const GPT_6_SOL_MODEL = "gpt-6-sol";
+const GPT_6_LUNA_MODEL = "gpt-6-luna";
+const GPT_6_SOL_EFFORTS = GPT_6_ASTRA_EFFORTS;
+const GPT_6_LUNA_EFFORTS = [
+    "low",
+    "medium",
+    "high",
+    "xhigh",
+    "max",
+];
 /**
  * Cyber-specialty models from the upstream Codex catalog
  * (openai/codex `codex-rs/models-manager/models.json`).
@@ -166,33 +198,43 @@ const GPT_5_5_RELEASE_MODEL = "gpt-5.5-2026-04-23";
 const GPT_5_5_PRO_RELEASE_MODEL = "gpt-5.5-pro-2026-04-23";
 const GPT_5_5_RELEASE_COMPAT_MODEL = "gpt-5.5-20260423";
 const GPT_5_5_PRO_RELEASE_COMPAT_MODEL = "gpt-5.5-pro-20260423";
+/**
+ * Where an unrecognised `gpt-5.<minor>` id lands. Every minor below 5.5 is
+ * retired (OpenAI deprecations page, upstream catalog removals), so each maps
+ * to the replacement OpenAI names for it rather than to a model that no longer
+ * answers: 5.1/5.2 to Sol, 5.4 to GPT-6 Sol/Luna as upstream Codex migrates it,
+ * and every `pro` to the one live pro model.
+ */
 const GENERAL_GPT5_VERSION_CATALOG = {
     1: {
-        base: "gpt-5.1",
+        base: GPT_5_6_SOL_MODEL,
+        pro: GPT_5_5_PRO_CANONICAL_MODEL,
     },
     2: {
-        base: "gpt-5.2",
-        pro: "gpt-5.2-pro",
+        base: GPT_5_6_SOL_MODEL,
+        pro: GPT_5_5_PRO_CANONICAL_MODEL,
     },
     4: {
-        base: DEFAULT_MODEL,
-        pro: "gpt-5.4-pro",
-        mini: "gpt-5.4-mini",
-        nano: "gpt-5.4-nano",
+        base: GPT_6_SOL_MODEL,
+        pro: GPT_5_5_PRO_CANONICAL_MODEL,
+        mini: GPT_6_LUNA_MODEL,
+        nano: GPT_6_LUNA_MODEL,
     },
     5: {
         base: GPT_5_5_CANONICAL_MODEL,
         pro: GPT_5_5_PRO_CANONICAL_MODEL,
-        mini: "gpt-5-mini",
-        nano: "gpt-5-nano",
+        mini: GPT_5_6_TERRA_MODEL,
+        nano: GPT_5_6_LUNA_MODEL,
     },
 };
 const GENERAL_GPT5_STABLE_VARIANTS = GENERAL_GPT5_VERSION_CATALOG[5];
+// `gpt-5-mini`/`gpt-5-nano` point at snapshots OpenAI retires on 2026-12-11;
+// the deprecations page names Terra and Luna as their replacements.
 const GENERAL_GPT5_GENERIC_VARIANTS = {
     base: DEFAULT_MODEL,
     pro: GPT_5_5_PRO_CANONICAL_MODEL,
-    mini: "gpt-5-mini",
-    nano: "gpt-5-nano",
+    mini: GPT_5_6_TERRA_MODEL,
+    nano: GPT_5_6_LUNA_MODEL,
 };
 /**
  * Effective model profiles keyed by canonical model name.
@@ -203,41 +245,6 @@ const GENERAL_GPT5_GENERIC_VARIANTS = {
  * present in the latest upstream release.
  */
 export const MODEL_PROFILES = {
-    [CURRENT_CODEX_MODEL]: {
-        normalizedModel: CURRENT_CODEX_MODEL,
-        promptFamily: "gpt-5-codex",
-        defaultReasoningEffort: "high",
-        supportedReasoningEfforts: ["low", "medium", "high", "xhigh"],
-        capabilities: TOOL_CAPABILITIES.basic,
-    },
-    "gpt-5.4": {
-        normalizedModel: "gpt-5.4",
-        promptFamily: "gpt-5.2",
-        defaultReasoningEffort: "none",
-        supportedReasoningEfforts: ["none", "low", "medium", "high", "xhigh"],
-        capabilities: TOOL_CAPABILITIES.full,
-    },
-    "gpt-5.4-pro": {
-        normalizedModel: "gpt-5.4-pro",
-        promptFamily: "gpt-5.2",
-        defaultReasoningEffort: "high",
-        supportedReasoningEfforts: ["medium", "high", "xhigh"],
-        capabilities: TOOL_CAPABILITIES.computerAndCompact,
-    },
-    "gpt-5.4-mini": {
-        normalizedModel: "gpt-5.4-mini",
-        promptFamily: "gpt-5.2",
-        defaultReasoningEffort: "medium",
-        supportedReasoningEfforts: ["medium"],
-        capabilities: TOOL_CAPABILITIES.compactOnly,
-    },
-    "gpt-5.4-nano": {
-        normalizedModel: "gpt-5.4-nano",
-        promptFamily: "gpt-5.2",
-        defaultReasoningEffort: "medium",
-        supportedReasoningEfforts: ["medium"],
-        capabilities: TOOL_CAPABILITIES.compactOnly,
-    },
     // Like GPT-5.6, GPT-6 Astra ships its base instructions inline in the
     // upstream model catalog rather than as a `gpt_6_prompt.md`, so it stays on
     // the GPT-5.2 prompt family with every other post-5.2 general model. Adding
@@ -258,6 +265,20 @@ export const MODEL_PROFILES = {
         supportedReasoningEfforts: GPT_6_ASTRA_EFFORTS,
         capabilities: TOOL_CAPABILITIES.full,
     },
+    [GPT_6_SOL_MODEL]: {
+        normalizedModel: GPT_6_SOL_MODEL,
+        promptFamily: "gpt-5.2",
+        defaultReasoningEffort: "medium",
+        supportedReasoningEfforts: GPT_6_SOL_EFFORTS,
+        capabilities: TOOL_CAPABILITIES.full,
+    },
+    [GPT_6_LUNA_MODEL]: {
+        normalizedModel: GPT_6_LUNA_MODEL,
+        promptFamily: "gpt-5.2",
+        defaultReasoningEffort: "medium",
+        supportedReasoningEfforts: GPT_6_LUNA_EFFORTS,
+        capabilities: TOOL_CAPABILITIES.full,
+    },
     [DAYBREAK_BLUE_MODEL]: {
         normalizedModel: DAYBREAK_BLUE_MODEL,
         promptFamily: "gpt-5.2",
@@ -310,41 +331,12 @@ export const MODEL_PROFILES = {
         supportedReasoningEfforts: ["medium", "high", "xhigh"],
         capabilities: TOOL_CAPABILITIES.computerAndCompact,
     },
-    "gpt-5.2-pro": {
-        normalizedModel: "gpt-5.2-pro",
-        promptFamily: "gpt-5.2",
-        defaultReasoningEffort: "high",
-        supportedReasoningEfforts: ["medium", "high", "xhigh"],
-        capabilities: TOOL_CAPABILITIES.basic,
-    },
-    "gpt-5.2": {
-        normalizedModel: "gpt-5.2",
-        promptFamily: "gpt-5.2",
-        defaultReasoningEffort: "none",
-        supportedReasoningEfforts: ["none", "low", "medium", "high", "xhigh"],
-        capabilities: TOOL_CAPABILITIES.basic,
-    },
-    "gpt-5.1": {
-        normalizedModel: "gpt-5.1",
-        promptFamily: "gpt-5.1",
-        defaultReasoningEffort: "none",
-        supportedReasoningEfforts: ["none", "low", "medium", "high"],
-        capabilities: TOOL_CAPABILITIES.basic,
-    },
-    "gpt-5-mini": {
-        normalizedModel: "gpt-5-mini",
-        promptFamily: "gpt-5.2",
-        defaultReasoningEffort: "medium",
-        supportedReasoningEfforts: ["medium"],
-        capabilities: TOOL_CAPABILITIES.compactOnly,
-    },
-    "gpt-5-nano": {
-        normalizedModel: "gpt-5-nano",
-        promptFamily: "gpt-5.2",
-        defaultReasoningEffort: "medium",
-        supportedReasoningEfforts: ["medium"],
-        capabilities: TOOL_CAPABILITIES.compactOnly,
-    },
+    // Every model older than 5.5 is retired: `gpt-5.1`, `gpt-5.2` and the whole
+    // 5.4 family left the upstream Codex catalog, the codex models and the
+    // chat-latest snapshots are past their shutdown date on OpenAI's
+    // deprecations page, and `gpt-5-mini`/`gpt-5-nano` shut down 2026-12-11.
+    // Their ids stay accepted as aliases of the named replacement (below) so an
+    // old config keeps working, but no request is sent under a dead name.
 };
 const MODEL_MAP = {};
 function addAlias(alias, normalizedModel) {
@@ -382,6 +374,11 @@ function addGpt6Aliases() {
     // rather than letting it fall through to GPT-5.5.
     addEffortAliases("astra", GPT_6_ASTRA_MODEL, GPT_6_ASTRA_EFFORTS);
     addEffortAliases("astra-aeon", GPT_6_ASTRA_AEON_MODEL, GPT_6_ASTRA_EFFORTS);
+    // No bare `sol`/`luna` aliases: those names already mean the 5.6 tiers to
+    // anyone who used them before 2026-09-22, and re-pointing them would swap a
+    // user's model generation without asking.
+    addEffortAliases(GPT_6_SOL_MODEL, GPT_6_SOL_MODEL, GPT_6_SOL_EFFORTS);
+    addEffortAliases(GPT_6_LUNA_MODEL, GPT_6_LUNA_MODEL, GPT_6_LUNA_EFFORTS);
 }
 function addDaybreakAliases() {
     addEffortAliases(DAYBREAK_BLUE_MODEL, DAYBREAK_BLUE_MODEL, DAYBREAK_EFFORTS);
@@ -396,35 +393,61 @@ function addGeneralAliases() {
     addReasoningAliases(GPT_5_5_PRO_CANONICAL_MODEL, GPT_5_5_PRO_CANONICAL_MODEL);
     addReasoningAliases(GPT_5_5_PRO_RELEASE_MODEL, GPT_5_5_PRO_CANONICAL_MODEL);
     addReasoningAliases(GPT_5_5_PRO_RELEASE_COMPAT_MODEL, GPT_5_5_PRO_CANONICAL_MODEL);
-    addReasoningAliases("gpt-5.4", "gpt-5.4");
-    addReasoningAliases("gpt-5.4-pro", "gpt-5.4-pro");
-    addReasoningAliases("gpt-5.4-mini", "gpt-5.4-mini");
-    addReasoningAliases("gpt-5.4-nano", "gpt-5.4-nano");
-    addReasoningAliases("gpt-5.2-pro", "gpt-5.2-pro");
     addReasoningAliases("gpt-5-pro", GPT_5_5_PRO_CANONICAL_MODEL);
-    addReasoningAliases("gpt-5.2", "gpt-5.2");
-    addReasoningAliases("gpt-5.1", "gpt-5.1");
     addReasoningAliases("gpt-5", DEFAULT_MODEL);
-    addReasoningAliases("gpt-5-mini", "gpt-5-mini");
-    addReasoningAliases("gpt-5-nano", "gpt-5-nano");
-    addReasoningAliases("gpt-5.1-chat-latest", "gpt-5.1");
-    addReasoningAliases("gpt-5-chat-latest", DEFAULT_MODEL);
 }
-function addCodexAliases() {
-    addReasoningAliases(CURRENT_CODEX_MODEL, CURRENT_CODEX_MODEL);
-    addReasoningAliases("gpt-5.3-codex-spark", CURRENT_CODEX_MODEL);
-    addReasoningAliases(LEGACY_CODEX_MODEL, CURRENT_CODEX_MODEL);
-    addReasoningAliases("gpt-5.2-codex", CURRENT_CODEX_MODEL);
-    addReasoningAliases("gpt-5.1-codex", CURRENT_CODEX_MODEL);
+/**
+ * Retired general ids, each kept as an alias of the replacement OpenAI names
+ * for it: the upstream Codex catalog's migration target where it has one
+ * (`gpt-5.4` -> GPT-6 Sol, `gpt-5.4-mini` -> GPT-6 Luna), otherwise the
+ * deprecations page's recommended replacement. The effort-suffixed forms
+ * (`gpt-5.4-high`, `gpt-5.1-none`, ...) come along so old configs resolve;
+ * efforts the replacement does not accept are coerced as for any request.
+ */
+const RETIRED_GENERAL_MODEL_REPLACEMENTS = {
+    "gpt-5.4": GPT_6_SOL_MODEL,
+    "gpt-5.4-mini": GPT_6_LUNA_MODEL,
+    "gpt-5.4-nano": GPT_6_LUNA_MODEL,
+    "gpt-5.4-pro": GPT_5_5_PRO_CANONICAL_MODEL,
+    "gpt-5.2": GPT_5_6_SOL_MODEL,
+    "gpt-5.2-pro": GPT_5_5_PRO_CANONICAL_MODEL,
+    "gpt-5.1": GPT_5_6_SOL_MODEL,
+    "gpt-5-mini": GPT_5_6_TERRA_MODEL,
+    "gpt-5-nano": GPT_5_6_LUNA_MODEL,
+    "gpt-5-chat-latest": GPT_5_6_SOL_MODEL,
+    "gpt-5.1-chat-latest": GPT_5_6_SOL_MODEL,
+    "gpt-5.2-chat-latest": GPT_5_6_SOL_MODEL,
+    "gpt-5.3-chat-latest": GPT_5_6_SOL_MODEL,
+};
+/**
+ * Retired codex ids. All shut down on OpenAI's deprecations page or left the
+ * upstream Codex catalog; OpenAI names `gpt-5.6-sol` as the replacement for
+ * the full-size ones and `gpt-5.6-terra` for `gpt-5.1-codex-mini`.
+ */
+const RETIRED_CODEX_MODEL_REPLACEMENTS = {
+    "gpt-5.3-codex": CURRENT_CODEX_MODEL,
+    "gpt-5.3-codex-spark": CURRENT_CODEX_MODEL,
+    "gpt-5.2-codex": CURRENT_CODEX_MODEL,
+    "gpt-5.1-codex": CURRENT_CODEX_MODEL,
+    "gpt-5.1-codex-max": CURRENT_CODEX_MODEL,
+    "gpt-5-codex": CURRENT_CODEX_MODEL,
+    "codex-max": CURRENT_CODEX_MODEL,
+    "gpt-5.1-codex-mini": CODEX_MINI_REPLACEMENT_MODEL,
+    "gpt-5-codex-mini": CODEX_MINI_REPLACEMENT_MODEL,
+    "codex-mini-latest": CODEX_MINI_REPLACEMENT_MODEL,
+};
+/** Every retired id and its replacement, for diagnostics and tests. */
+export const RETIRED_MODEL_REPLACEMENTS = {
+    ...RETIRED_GENERAL_MODEL_REPLACEMENTS,
+    ...RETIRED_CODEX_MODEL_REPLACEMENTS,
+};
+function addRetiredAliases() {
+    for (const [retired, replacement] of Object.entries(RETIRED_MODEL_REPLACEMENTS)) {
+        addReasoningAliases(retired, replacement);
+    }
     addAlias("gpt_5_codex", CURRENT_CODEX_MODEL);
-    addReasoningAliases("codex-max", CURRENT_CODEX_MODEL);
-    addReasoningAliases("gpt-5.1-codex-max", CURRENT_CODEX_MODEL);
-    addAlias("codex-max", CURRENT_CODEX_MODEL);
-    addAlias("codex-mini-latest", CURRENT_CODEX_MODEL);
-    addReasoningAliases("gpt-5-codex-mini", CURRENT_CODEX_MODEL);
-    addReasoningAliases("gpt-5.1-codex-mini", CURRENT_CODEX_MODEL);
 }
-addCodexAliases();
+addRetiredAliases();
 addGeneralAliases();
 addGpt56Aliases();
 addGpt6Aliases();
@@ -461,39 +484,21 @@ function resolveStableGeneralGpt5Variant(variant) {
     }
     throw new Error(`Stable GPT-5 fallback is missing for variant ${variant}`);
 }
+/**
+ * Any id carrying a `codex` token. Every codex model is retired, so this only
+ * picks the replacement: a `mini` codex id goes to Terra, everything else to
+ * Sol (see RETIRED_CODEX_MODEL_REPLACEMENTS).
+ */
 function resolveCodexCatalogModel(modelId) {
     const normalized = modelId.toLowerCase();
-    if (normalized.includes("gpt-5.3-codex-spark") ||
-        normalized.includes("gpt 5.3 codex spark")) {
-        return CURRENT_CODEX_MODEL;
-    }
-    if (normalized.includes("gpt-5.3-codex") ||
-        normalized.includes("gpt 5.3 codex")) {
-        return CURRENT_CODEX_MODEL;
-    }
-    if (normalized.includes("gpt-5.2-codex") ||
-        normalized.includes("gpt 5.2 codex")) {
-        return CURRENT_CODEX_MODEL;
-    }
-    if (normalized.includes("gpt-5.1-codex-max") ||
-        normalized.includes("gpt 5.1 codex max")) {
-        return CURRENT_CODEX_MODEL;
-    }
-    if (normalized.includes("gpt-5.1-codex-mini") ||
-        normalized.includes("gpt 5.1 codex mini") ||
-        normalized.includes("codex-mini-latest") ||
-        normalized.includes("gpt-5-codex-mini") ||
-        normalized.includes("gpt 5 codex mini")) {
-        return CURRENT_CODEX_MODEL;
+    if (!normalized.includes("codex")) {
+        return undefined;
     }
-    if (normalized.includes("gpt-5-codex") ||
-        normalized.includes("gpt 5 codex") ||
-        normalized.includes("gpt-5.1-codex") ||
-        normalized.includes("gpt 5.1 codex") ||
-        normalized.includes("codex")) {
-        return CURRENT_CODEX_MODEL;
+    // Anchored so `codex-minimal` (an effort suffix) is not read as a mini.
+    if (/codex[- ]mini(?!mal)/.test(normalized)) {
+        return CODEX_MINI_REPLACEMENT_MODEL;
     }
-    return undefined;
+    return CURRENT_CODEX_MODEL;
 }
 /**
  * Resolve GPT-5.6 identifiers that are not exact aliases (for example a future
@@ -514,8 +519,9 @@ function resolveCodexCatalogModel(modelId) {
  * `gpt 5` token pair) and every unrecognised GPT-6 id lands on `DEFAULT_MODEL`,
  * running GPT-5.5 for a caller who asked for the frontier model. `aeon` keeps
  * its own canonical id because it is a behaviourally different model (long
- * horizon), not a rename of the flagship; everything else resolves to the
- * flagship, matching OpenAI's bare `gpt-6` alias.
+ * horizon), not a rename of the flagship. A `sol`/`luna` tier token picks that
+ * tier; everything else resolves to the flagship, matching OpenAI's bare
+ * `gpt-6` alias.
  *
  * Ids carrying a `codex` token are left to `resolveCodexCatalogModel`, exactly
  * as the 5.6 resolver defers them.
@@ -547,6 +553,16 @@ function resolveGpt6CatalogModel(modelId) {
     }
     if (tokens.includes("aeon"))
         return GPT_6_ASTRA_AEON_MODEL;
+    if (isAstra)
+        return GPT_6_ASTRA_MODEL;
+    // Before Sol and Luna existed every non-Astra GPT-6 id fell to the line
+    // below, so `gpt-6-luna` silently ran Astra at 100x Luna's price. `terra`
+    // goes to Sol because there is no GPT-6 Terra and upstream migrates
+    // `gpt-5.6-terra` users to Sol.
+    if (tokens.includes("luna"))
+        return GPT_6_LUNA_MODEL;
+    if (tokens.includes("sol") || tokens.includes("terra"))
+        return GPT_6_SOL_MODEL;
     return GPT_6_ASTRA_MODEL;
 }
 /**
@@ -647,6 +663,17 @@ export function resolveNormalizedModel(model) {
     if (mappedModel) {
         return mappedModel;
     }
+    // `max`/`ultra` are not generated as aliases for the pre-5.6 ids, so
+    // `gpt-5.3-chat-latest-max` missed the retired table and the general GPT-5
+    // resolver sent it to 5.5 instead of its replacement. Retry the lookup
+    // without the suffix before any fuzzy resolver runs.
+    const withoutTopEffort = modelId.replace(/-(max|ultra)$/i, "");
+    if (withoutTopEffort !== modelId) {
+        const mappedWithoutEffort = lookupMappedModel(withoutTopEffort);
+        if (mappedWithoutEffort) {
+            return mappedWithoutEffort;
+        }
+    }
     // Daybreak first: its slugs carry neither a `codex` nor a `gpt 5` token, so
     // every other resolver declines them and they would reach DEFAULT_MODEL.
     const daybreakCatalogModel = resolveDaybreakCatalogModel(modelId);

package/dist/lib/runtime/config-toml.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"config-toml.d.ts","sourceRoot":"","sources":["../../../lib/runtime/config-toml.ts"],"names":[],"mappings":"AAEA,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAqBvD;AA4FD,wBAAgB,4BAA4B,CAC3C,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,MAAM,GACpB,MAAM,CAiDR;AAED,wBAAgB,8BAA8B,CAC7C,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,MAAM,GACpB,MAAM,CAkDR;AA2BD,wBAAgB,2CAA2C,CAC1D,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,MAAM,EACf,YAAY,SAAK,GACf,MAAM,CAkBR;AAED,wBAAgB,4CAA4C,CAC3D,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,MAAM,GACpB,MAAM,CAYR;AAED;;;;;;;GAOG;AACH,wBAAgB,gCAAgC,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAoB3E;AAED;;;;;;;GAOG;AACH,wBAAgB,yDAAyD,CACxE,aAAa,EAAE,MAAM,EACrB,eAAe,SAAW,GACxB,MAAM,CAkBR"}
+{"version":3,"file":"config-toml.d.ts","sourceRoot":"","sources":["../../../lib/runtime/config-toml.ts"],"names":[],"mappings":"AAEA,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAqBvD;AA4FD,wBAAgB,4BAA4B,CAC3C,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,MAAM,GACpB,MAAM,CAiDR;AAED,wBAAgB,8BAA8B,CAC7C,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,MAAM,GACpB,MAAM,CAkDR;AA4BD,wBAAgB,2CAA2C,CAC1D,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,MAAM,EACf,YAAY,SAAK,GACf,MAAM,CAkBR;AAED,wBAAgB,4CAA4C,CAC3D,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,MAAM,GACpB,MAAM,CAYR;AAED;;;;;;;GAOG;AACH,wBAAgB,gCAAgC,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAoB3E;AAED;;;;;;;GAOG;AACH,wBAAgB,yDAAyD,CACxE,aAAa,EAAE,MAAM,EACrB,eAAe,SAAW,GACxB,MAAM,CAkBR"}

package/dist/lib/runtime/config-toml.js

@@ -202,6 +202,7 @@ function createRuntimeRotationProviderBlock(baseUrl, clientApiKey = "") {
         'name = "codex-multi-auth"',
         `base_url = ${tomlStringLiteral(baseUrl)}`,
         "requires_openai_auth = false",
+        'http_headers = { "x-openai-actor-authorization" = "codex-multi-auth-local" }',
         'wire_api = "responses"',
     ];
     if (clientApiKey.trim().length > 0) {

package/dist/lib/runtime-rotation-proxy.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"runtime-rotation-proxy.d.ts","sourceRoot":"","sources":["../../lib/runtime-rotation-proxy.ts"],"names":[],"mappings":"AAGA,OAAO,EAIN,KAAK,cAAc,EACnB,MAAM,eAAe,CAAC;AAkCvB,OAAO,EAAkB,KAAK,WAAW,EAAE,MAAM,oBAAoB,CAAC;AAqDtE,OAAO,KAAK,EAKX,2BAA2B,EAC3B,0BAA0B,EAE1B,MAAM,oCAAoC,CAAC;AAe5C,YAAY,EACX,2BAA2B,EAC3B,0BAA0B,EAC1B,0BAA0B,GAC1B,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EACN,+BAA+B,EAC/B,0BAA0B,GAC1B,MAAM,kCAAkC,CAAC;AAC1C,YAAY,EAAE,0BAA0B,EAAE,MAAM,kCAAkC,CAAC;AACnF,OAAO,EAAE,aAAa,EAAE,MAAM,yCAAyC,CAAC;AACxE,OAAO,EACN,+BAA+B,EAC/B,8BAA8B,EAC9B,uBAAuB,EACvB,uBAAuB,GACvB,MAAM,oCAAoC,CAAC;AAC5C,YAAY,EAAE,WAAW,EAAE,MAAM,oCAAoC,CAAC;AA8DtE,kFAAkF;AAClF;;;;;;;GAOG;AACH,wBAAgB,+BAA+B,CAC9C,OAAO,EAAE,IAAI,CAAC,cAAc,EAAE,WAAW,GAAG,OAAO,GAAG,cAAc,CAAC,GAAG;IACvE,gFAAgF;IAChF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;CAClB,GACC,MAAM,CAmBR;AAED,wBAAgB,qBAAqB,CACpC,OAAO,EAAE,IAAI,CAAC,cAAc,EAAE,WAAW,GAAG,OAAO,GAAG,cAAc,CAAC,GAAG;IACvE,gFAAgF;IAChF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;CAClB,EACD,MAAM,EAAE,WAAW,EACnB,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GACnB,MAAM,CAER;AA0ED;;;;;;;GAOG;AACH,wBAAgB,6BAA6B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAKpE;AAED,wBAAgB,2BAA2B,CAC1C,GAAG,GAAE,MAAM,CAAC,UAAwB,GAClC,MAAM,CAQR;AAueD;;;;;;;;GAQG;AACH,wBAAgB,2BAA2B,CAC1C,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,GAAG,SAAS,GACvC,MAAM,GAAG,IAAI,CAUf;AAED,wBAAsB,yBAAyB,CAC9C,OAAO,EAAE,2BAA2B,GAClC,OAAO,CAAC,0BAA0B,CAAC,CA6LrC"}
+{"version":3,"file":"runtime-rotation-proxy.d.ts","sourceRoot":"","sources":["../../lib/runtime-rotation-proxy.ts"],"names":[],"mappings":"AAGA,OAAO,EAIN,KAAK,cAAc,EACnB,MAAM,eAAe,CAAC;AAkCvB,OAAO,EAAkB,KAAK,WAAW,EAAE,MAAM,oBAAoB,CAAC;AAqDtE,OAAO,KAAK,EAKX,2BAA2B,EAC3B,0BAA0B,EAE1B,MAAM,oCAAoC,CAAC;AAe5C,YAAY,EACX,2BAA2B,EAC3B,0BAA0B,EAC1B,0BAA0B,GAC1B,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EACN,+BAA+B,EAC/B,0BAA0B,GAC1B,MAAM,kCAAkC,CAAC;AAC1C,YAAY,EAAE,0BAA0B,EAAE,MAAM,kCAAkC,CAAC;AACnF,OAAO,EAAE,aAAa,EAAE,MAAM,yCAAyC,CAAC;AACxE,OAAO,EACN,+BAA+B,EAC/B,8BAA8B,EAC9B,uBAAuB,EACvB,uBAAuB,GACvB,MAAM,oCAAoC,CAAC;AAC5C,YAAY,EAAE,WAAW,EAAE,MAAM,oCAAoC,CAAC;AA8DtE,kFAAkF;AAClF;;;;;;;GAOG;AACH,wBAAgB,+BAA+B,CAC9C,OAAO,EAAE,IAAI,CAAC,cAAc,EAAE,WAAW,GAAG,OAAO,GAAG,cAAc,CAAC,GAAG;IACvE,gFAAgF;IAChF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;CAClB,GACC,MAAM,CAmBR;AAED,wBAAgB,qBAAqB,CACpC,OAAO,EAAE,IAAI,CAAC,cAAc,EAAE,WAAW,GAAG,OAAO,GAAG,cAAc,CAAC,GAAG;IACvE,gFAAgF;IAChF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;CAClB,EACD,MAAM,EAAE,WAAW,EACnB,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GACnB,MAAM,CAER;AA0ED;;;;;;;GAOG;AACH,wBAAgB,6BAA6B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAKpE;AAED,wBAAgB,2BAA2B,CAC1C,GAAG,GAAE,MAAM,CAAC,UAAwB,GAClC,MAAM,CAQR;AA0fD;;;;;;;;GAQG;AACH,wBAAgB,2BAA2B,CAC1C,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,GAAG,SAAS,GACvC,MAAM,GAAG,IAAI,CAUf;AAED,wBAAsB,yBAAyB,CAC9C,OAAO,EAAE,2BAA2B,GAClC,OAAO,CAAC,0BAA0B,CAAC,CA6LrC"}

package/dist/lib/runtime-rotation-proxy.js

@@ -208,6 +208,12 @@ const ALLOWED_MODELS_PATHS = new Set([
     URL_PATHS.MODELS,
     `/v1${URL_PATHS.MODELS}`,
 ]);
+const ALLOWED_IMAGE_PATHS = new Set([
+    "/images/generations",
+    "/images/edits",
+    "/v1/images/generations",
+    "/v1/images/edits",
+]);
 const ALLOWED_THREAD_GOAL_PATHS = new Set([
     "/thread/goal/get",
     "/thread/goal/set",
@@ -252,6 +258,8 @@ function createOutboundHeaders(incoming, account, accessToken, accountId) {
     // header would ride along with the managed OAuth Bearer to OpenAI.
     headers.delete("cookie");
     headers.delete("proxy-authorization");
+    // Local capability marker for Codex image_gen, never an upstream credential.
+    headers.delete("x-openai-actor-authorization");
     headers.set("authorization", `Bearer ${accessToken}`);
     headers.set(OPENAI_HEADERS.ACCOUNT_ID, accountId);
     headers.set(OPENAI_HEADERS.BETA, OPENAI_HEADER_VALUES.BETA_RESPONSES);
@@ -488,16 +496,22 @@ function buildResponsesRequestContext(req, body) {
         method: "POST",
         upstreamPath: URL_PATHS.CODEX_RESPONSES,
         model,
-        // The /codex/responses path is codex-family. A model-less request must
-        // bucket into the codex family (rotation/cooldown/budget), so fall back to
-        // the current codex model — NOT the general DEFAULT_MODEL (gpt-5.5), whose
-        // family is gpt-5.2 and would mis-account a pass-through codex request.
+        // A model-less request buckets under the model a codex request runs on.
+        // With every codex model retired that is `gpt-5.6-sol`, whose family is
+        // gpt-5.2, the same family as DEFAULT_MODEL.
         family: getModelFamily(model ?? CURRENT_CODEX_MODEL),
         stream: parsedBody?.stream === true,
         sessionKey: resolveSessionKey(headers, parsedBody),
         stableSessionKey: resolveStableSessionKey(headers, parsedBody),
     };
 }
+function buildImageRequestContext(req, body, pathname) {
+    return {
+        ...buildResponsesRequestContext(req, body),
+        upstreamPath: `/codex${pathname.replace(/^\/v1/, "")}`,
+        family: "codex",
+    };
+}
 function buildModelsRequestContext(req) {
     return {
         body: Buffer.alloc(0),
@@ -555,7 +569,7 @@ function writeJson(res, status, payload) {
 function writeMethodOrPathError(res) {
     writeJson(res, 404, {
         error: {
-            message: "Runtime rotation proxy only accepts Responses API, model discovery, and Codex thread goal requests.",
+            message: "Runtime rotation proxy only accepts Responses API, images, model discovery, and Codex thread goal requests.",
             code: "runtime_rotation_proxy_not_found",
         },
     });
@@ -803,21 +817,26 @@ async function handleRequestInner(state, req, res, traceId) {
         }
         const isResponsesRequest = req.method === "POST" && isResponsesPath(incomingUrl.pathname);
         const isModelsRequest = req.method === "GET" && isModelsPath(incomingUrl.pathname);
+        const isImageRequest = req.method === "POST" && ALLOWED_IMAGE_PATHS.has(incomingUrl.pathname);
         const isThreadGoalRequest = (req.method === "GET" || req.method === "POST") &&
             isThreadGoalPath(incomingUrl.pathname);
-        if (!isResponsesRequest && !isModelsRequest && !isThreadGoalRequest) {
+        if (!isResponsesRequest && !isModelsRequest &&
+            !isThreadGoalRequest && !isImageRequest) {
             writeMethodOrPathError(res);
             return;
         }
         state.status.totalRequests += 1;
-        const requestBody = isResponsesRequest || (isThreadGoalRequest && req.method === "POST")
+        const requestBody = isResponsesRequest || isImageRequest ||
+            (isThreadGoalRequest && req.method === "POST")
             ? await readRequestBody(req, state.maxRequestBodyBytes)
             : Buffer.alloc(0);
         const context = isModelsRequest
             ? buildModelsRequestContext(req)
             : isThreadGoalRequest
                 ? buildThreadGoalRequestContext(req, requestBody, incomingUrl.pathname)
-                : buildResponsesRequestContext(req, requestBody);
+                : isImageRequest
+                    ? buildImageRequestContext(req, requestBody, incomingUrl.pathname)
+                    : buildResponsesRequestContext(req, requestBody);
         const requestStartedAt = state.now();
         let policyDecision = null;
         let projectKey = null;
@@ -848,7 +867,9 @@ async function handleRequestInner(state, req, res, traceId) {
                 ? "models"
                 : isThreadGoalRequest
                     ? "thread-goal"
-                    : "responses",
+                    : isImageRequest
+                        ? "images"
+                        : "responses",
             model: context.model,
             projectKey,
             requestId: traceId,
@@ -1287,9 +1308,27 @@ async function handleRequestInner(state, req, res, traceId) {
             recordLastRuntimeAccount(state.status, accountIdentity);
             const outboundHeaders = createOutboundHeaders(context.headers, refreshed.account, refreshed.accessToken, accountId);
             let upstream;
+            // Abort the in-flight upstream fetch when the client disconnects
+            // before headers arrive. Image generation holds upstream capacity
+            // for the full fetch timeout, so a caller that goes away right
+            // after sending must not leave that work running. `forwardStreamingResponse`
+            // already cancels the stream once headers are written; this covers
+            // the pre-header window instead. `writableEnded` distinguishes a
+            // premature close from the clean `res.end()` that ends every request.
+            //
+            // The listener is removed in `finally` once the fetch settles. Without
+            // that, every retry above the 10-listener default threshold that reaches
+            // this fetch leaks another one-shot `close` handler onto the same `res`,
+            // emitting `MaxListenersExceededWarning` when `retryAllAccountsMaxRetries`
+            // is high and the account pool is large.
+            const fetchAbortController = new AbortController();
+            const onClientClose = () => {
+                if (!res.writableEnded)
+                    fetchAbortController.abort();
+            };
             try {
                 state.status.upstreamRequests += 1;
-                const fetchAbortController = new AbortController();
+                res.once("close", onClientClose);
                 const upstreamRequestInit = {
                     method: context.method,
                     headers: outboundHeaders,
@@ -1298,7 +1337,10 @@ async function handleRequestInner(state, req, res, traceId) {
                 if (context.method === "POST") {
                     upstreamRequestInit.body = context.body;
                 }
-                upstream = await withTimeout(state.fetchImpl(upstreamUrl, upstreamRequestInit), state.fetchTimeoutMs, () => fetchAbortController.abort(), `upstream fetch timed out after ${state.fetchTimeoutMs}ms`);
+                const fetchTimeoutMs = isImageRequest
+                    ? Math.max(state.fetchTimeoutMs, 300_000)
+                    : state.fetchTimeoutMs;
+                upstream = await withTimeout(state.fetchImpl(upstreamUrl, upstreamRequestInit), fetchTimeoutMs, () => fetchAbortController.abort(), `upstream fetch timed out after ${fetchTimeoutMs}ms`);
             }
             catch (error) {
                 // errors-logging-08: a custom fetchImpl, a proxy agent, or an undici
@@ -1315,6 +1357,20 @@ async function handleRequestInner(state, req, res, traceId) {
                     error: transportError,
                 });
                 refundConsumedPoolToken(refreshed.account);
+                // A timeout may occur after generation; do not retry within this request.
+                if (isImageRequest) {
+                    writeJson(res, 502, {
+                        error: {
+                            code: "image_upstream_transport_error",
+                            message: "Image upstream transport failed; not retried by the proxy.",
+                        },
+                    });
+                    await usageRecorder.record({
+                        outcome: "failure", statusCode: 502,
+                        errorCode: "image_upstream_transport_error", account: refreshed.account,
+                    });
+                    return;
+                }
                 // A pre-header transport exception is a property of the network path,
                 // not of this account's credentials or quota, so it must NOT feed the
                 // account's circuit breaker / health tracker: that is what would
@@ -1337,6 +1393,9 @@ async function handleRequestInner(state, req, res, traceId) {
                 noteRotation();
                 continue;
             }
+            finally {
+                res.off("close", onClientClose);
+            }
             reconcileManualSelection();
             const quotaSnapshot = readQuotaSchedulerSnapshot(upstream.headers, upstream.status, state.now());
             if (quotaSnapshot) {
@@ -1507,6 +1566,17 @@ async function handleRequestInner(state, req, res, traceId) {
                 noteRotation();
                 continue;
             }
+            // Forward image validation/moderation/5xx errors without cross-account replay.
+            // Explicit 429 quota and 401 auth handling above still use the existing pool.
+            if (isImageRequest && upstream.status >= 400) {
+                const forwarded = await forwardStreamingResponse(upstream, res, state.status, () => undefined, state.streamStallTimeoutMs);
+                await usageRecorder.record({
+                    outcome: "failure", statusCode: upstream.status,
+                    errorCode: forwarded ? "image_upstream_error" : "stream_forward_failed",
+                    account: refreshed.account,
+                });
+                return;
+            }
             if (upstream.status >= 500) {
                 const bodyText = await readErrorBody(upstream, state.streamStallTimeoutMs);
                 // A capacity 5xx is upstream load, not a broken account, so it

package/dist/lib/usage/ledger.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"ledger.d.ts","sourceRoot":"","sources":["../../../lib/usage/ledger.ts"],"names":[],"mappings":"AASA,OAAO,EAEN,KAAK,sBAAsB,EAE3B,KAAK,gBAAgB,EACrB,KAAK,gBAAgB,EAErB,KAAK,cAAc,EAEnB,KAAK,YAAY,EAGjB,KAAK,iBAAiB,EAEtB,MAAM,YAAY,CAAC;AAqOpB,wBAAgB,mBAAmB,IAAI,gBAAgB,CAMtD;AAED,wBAAsB,oBAAoB,CACzC,KAAK,EAAE,sBAAsB,GAC3B,OAAO,CAAC,cAAc,CAAC,CASzB;AAuKD,wBAAsB,mBAAmB,CACxC,KAAK,GAAE,gBAAqB,GAC1B,OAAO,CAAC,cAAc,EAAE,CAAC,CAc3B;AAuDD,wBAAgB,kBAAkB,CACjC,IAAI,EAAE,cAAc,EAAE,EACtB,KAAK,GAAE,iBAAsB,GAC3B,YAAY,CAwBd;AAED,wBAAsB,oBAAoB,CACzC,KAAK,GAAE,iBAAsB,GAC3B,OAAO,CAAC,YAAY,CAAC,CAEvB;AAED,wBAAsB,iBAAiB,CAAC,OAAO,GAAE;IAChD,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,iBAAiB,CAAC,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CA2B9B;AAED,wBAAgB,6BAA6B,IAAI,IAAI,CAEpD"}
+{"version":3,"file":"ledger.d.ts","sourceRoot":"","sources":["../../../lib/usage/ledger.ts"],"names":[],"mappings":"AASA,OAAO,EAEN,KAAK,sBAAsB,EAE3B,KAAK,gBAAgB,EACrB,KAAK,gBAAgB,EAErB,KAAK,cAAc,EAEnB,KAAK,YAAY,EAGjB,KAAK,iBAAiB,EAEtB,MAAM,YAAY,CAAC;AAsOpB,wBAAgB,mBAAmB,IAAI,gBAAgB,CAMtD;AAED,wBAAsB,oBAAoB,CACzC,KAAK,EAAE,sBAAsB,GAC3B,OAAO,CAAC,cAAc,CAAC,CASzB;AAuKD,wBAAsB,mBAAmB,CACxC,KAAK,GAAE,gBAAqB,GAC1B,OAAO,CAAC,cAAc,EAAE,CAAC,CAc3B;AAuDD,wBAAgB,kBAAkB,CACjC,IAAI,EAAE,cAAc,EAAE,EACtB,KAAK,GAAE,iBAAsB,GAC3B,YAAY,CAwBd;AAED,wBAAsB,oBAAoB,CACzC,KAAK,GAAE,iBAAsB,GAC3B,OAAO,CAAC,YAAY,CAAC,CAEvB;AAED,wBAAsB,iBAAiB,CAAC,OAAO,GAAE;IAChD,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,iBAAiB,CAAC,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CA2B9B;AAED,wBAAgB,6BAA6B,IAAI,IAAI,CAEpD"}