@prestyj/core 5.3.0 → 5.4.0

package/dist/model-registry.cjs

@@ -36,6 +36,7 @@ __export(model_registry_exports, {
   getAuthStorageKeys: () => getAuthStorageKeys,
   getContextWindow: () => getContextWindow,
   getDefaultModel: () => getDefaultModel,
+  getFastModel: () => getFastModel,
   getMaxThinkingLevel: () => getMaxThinkingLevel,
   getModel: () => getModel,
   getModelsForProvider: () => getModelsForProvider,
@@ -170,49 +171,63 @@ var MODELS = [
     maxThinkingLevel: "high"
   },
   // ── OpenAI (Codex) ─────────────────────────────────────
+  // GPT-5.6 family — three agentic coding tiers launched July 2026. All share a
+  // 372K context window: the Codex repo's models.json lists both `context_window`
+  // AND `max_context_window` as 372000 for every 5.6 variant, so (unlike 5.4/5.5,
+  // whose public-API 1M window exceeds the Codex-product cap) there's no split
+  // between the API and Codex-transport windows — a single `contextWindow` is
+  // correct in both code paths. All three take text+image input, freeform
+  // apply_patch, text+image web search, and parallel tool calls.
   {
-    id: "gpt-5.5",
-    name: "GPT-5.5",
+    // Sol — "Latest frontier agentic coding model." (priority 1, default low).
+    // Reasoning ladder: low → medium → high → xhigh → max → ultra. gg-ai's
+    // ThinkingLevel tops out at `max`, so that's the cap here; the `ultra` rung
+    // (auto task delegation) needs a ThinkingLevel extension to expose.
+    id: "gpt-5.6-sol",
+    name: "GPT-5.6 Sol",
     provider: "openai",
-    contextWindow: 105e4,
-    codexContextWindow: 272e3,
+    contextWindow: 372e3,
     maxOutputTokens: 128e3,
     supportsThinking: true,
     supportsImages: true,
     supportsVideo: false,
     costTier: "high",
-    maxThinkingLevel: "xhigh"
+    maxThinkingLevel: "max"
   },
   {
-    id: "gpt-5.4",
-    name: "GPT-5.4",
+    // Terra — "Balanced agentic coding model for everyday work." (priority 2,
+    // default medium). Same 372K context + ultra reasoning ladder as Sol.
+    id: "gpt-5.6-terra",
+    name: "GPT-5.6 Terra",
     provider: "openai",
-    contextWindow: 105e4,
-    codexContextWindow: 272e3,
+    contextWindow: 372e3,
     maxOutputTokens: 128e3,
     supportsThinking: true,
     supportsImages: true,
     supportsVideo: false,
-    costTier: "high",
-    maxThinkingLevel: "xhigh"
+    costTier: "medium",
+    maxThinkingLevel: "max"
   },
   {
-    id: "gpt-5.4-mini",
-    name: "GPT-5.4 Mini",
+    // Luna — "Fast and affordable agentic coding model." (priority 3, default
+    // medium). Same 372K context; reasoning tops out at `max` (no `ultra` rung).
+    id: "gpt-5.6-luna",
+    name: "GPT-5.6 Luna",
     provider: "openai",
-    contextWindow: 4e5,
+    contextWindow: 372e3,
     maxOutputTokens: 128e3,
     supportsThinking: true,
     supportsImages: true,
     supportsVideo: false,
     costTier: "low",
-    maxThinkingLevel: "xhigh"
+    maxThinkingLevel: "max"
   },
   {
-    id: "gpt-5.3-codex",
-    name: "GPT-5.3 Codex",
+    id: "gpt-5.5",
+    name: "GPT-5.5",
     provider: "openai",
-    contextWindow: 4e5,
+    contextWindow: 105e4,
+    codexContextWindow: 272e3,
     maxOutputTokens: 128e3,
     supportsThinking: true,
     supportsImages: true,
@@ -252,8 +267,8 @@ var MODELS = [
   },
   // ── Gemini ─────────────────────────────────────────────
   {
-    id: "gemini-3.1-flash-lite-preview",
-    name: "Gemini 3.1 Flash Lite Preview",
+    id: "gemini-3.1-flash-lite",
+    name: "Gemini 3.1 Flash Lite",
     provider: "gemini",
     contextWindow: 1048576,
     maxOutputTokens: 65536,
@@ -265,7 +280,10 @@ var MODELS = [
     maxThinkingLevel: "high"
   },
   {
-    id: "gemini-3.5-flash",
+    // Wire name `gemini-3-flash` — the Code Assist (OAuth) backend rejects the
+    // display string `gemini-3.5-flash` with a 404, so gemini-cli keeps this
+    // alternative name (SECONDARY_GEMINI_3_5_FLASH_MODEL) for that endpoint.
+    id: "gemini-3-flash",
     name: "Gemini 3.5 Flash",
     provider: "gemini",
     contextWindow: 1048576,
@@ -277,6 +295,22 @@ var MODELS = [
     costTier: "low",
     maxThinkingLevel: "high"
   },
+  {
+    // Gemini 3.1 Pro is public preview — gated behind Code Assist preview
+    // enablement, so free/personal OAuth accounts 404 on it (see
+    // ACCOUNT_GATED_MODELS in gg-ai's gemini provider).
+    id: "gemini-3.1-pro-preview",
+    name: "Gemini 3.1 Pro (Preview)",
+    provider: "gemini",
+    contextWindow: 1048576,
+    maxOutputTokens: 65536,
+    supportsThinking: true,
+    supportsImages: true,
+    supportsVideo: true,
+    maxVideoBytes: 20 * 1024 * 1024,
+    costTier: "high",
+    maxThinkingLevel: "high"
+  },
   // ── Moonshot (Kimi) ────────────────────────────────────
   {
     id: "kimi-k2.7-code",
@@ -467,8 +501,8 @@ function getVideoByteLimit(modelId) {
 }
 function getDefaultModel(provider) {
   if (provider === "xiaomi") return MODELS.find((m) => m.id === "mimo-v2.5-pro");
-  if (provider === "openai") return MODELS.find((m) => m.id === "gpt-5.5");
-  if (provider === "gemini") return MODELS.find((m) => m.id === "gemini-3.1-flash-lite-preview");
+  if (provider === "openai") return MODELS.find((m) => m.id === "gpt-5.6-sol");
+  if (provider === "gemini") return MODELS.find((m) => m.id === "gemini-3.1-flash-lite");
   if (provider === "glm") return MODELS.find((m) => m.id === "glm-5.2");
   if (provider === "moonshot") return MODELS.find((m) => m.id === "kimi-k2.7-code");
   if (provider === "minimax") return MODELS.find((m) => m.id === "MiniMax-M3");
@@ -501,6 +535,10 @@ function getSummaryModel(provider, currentModelId) {
   }
   return getModel(currentModelId) ?? getDefaultModel(provider);
 }
+function getFastModel(provider, currentModelId) {
+  const low = getModelsForProvider(provider).find((m) => m.costTier === "low");
+  return low ?? getModel(currentModelId) ?? getDefaultModel(provider);
+}
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   DEFAULT_MAX_VIDEO_BYTES,
@@ -509,6 +547,7 @@ function getSummaryModel(provider, currentModelId) {
   getAuthStorageKeys,
   getContextWindow,
   getDefaultModel,
+  getFastModel,
   getMaxThinkingLevel,
   getModel,
   getModelsForProvider,