@nodaro/shared 2.7.0 → 2.8.0

package/dist/index.cjs

@@ -4818,6 +4818,30 @@ var LLM_MODELS = [
     // direct is the reliability fallback only.
     directGeminiModel: "gemini-3.6-flash"
   },
+  {
+    id: "gemini-3.7-flash",
+    displayName: "Gemini 3.7 Flash",
+    desc: "Newest fast Gemini, agentic-tuned",
+    tier: "economy",
+    kieFormat: "chat-completions",
+    // KIE serves it on the OpenAI-compatible dialect under this slug
+    // (docs.kie.ai/market/gemini/gemini-3-7-flash-openai.md) — same
+    // chat-completions path shape as gemini-3.6-flash.
+    kieSlugOrModel: "gemini-3-7-flash-openai",
+    vendor: "google",
+    structuredOutputMode: "kie-response-format",
+    supportsImages: true,
+    // Google's own cap is 65,536, but the field feeds BOTH lanes and the KIE
+    // flash endpoints cap at 8192 (the measured 3.6 posture) — stay at the
+    // KIE-safe intersection, same reasoning as `reasoningEfforts` below.
+    maxOutputTokens: 8192,
+    // KIE's 3.7 endpoint enumerates reasoning_effort low | high (verified
+    // against its OpenAPI spec 2026-08-18), identical to 3.6.
+    reasoningEfforts: ["low", "high"],
+    // Assumed parity with 3.6 pending a live probe on the direct lane.
+    directReasoningEfforts: ["none", "low", "medium", "high"],
+    directGeminiModel: "gemini-3.7-flash"
+  },
   {
     id: "claude-haiku-4.5",
     displayName: "Claude Haiku 4.5",
@@ -4980,7 +5004,35 @@ var LLM_MODELS = [
     reasoningEfforts: ["none", "low", "medium", "high", "xhigh", "max"],
     supportsTemperature: false
   },
-  // grok-4.5 deferred — KIE chat endpoint not yet live (2026-07-13); add entry + rate row + docs when it activates.
+  {
+    id: "grok-4.6",
+    displayName: "Grok 4.6",
+    desc: "xAI flagship, strong reasoning",
+    tier: "standard",
+    kieFormat: "responses",
+    // KIE serves Grok on the responses dialect under its own family path —
+    // grok/v1/responses, NOT codex/v1/responses (llm-client derives the path
+    // from `vendor`). Live-verified end-to-end 2026-08-18: array `input`,
+    // `developer` system role, `input_image` URL vision, `text.format`
+    // json_schema enforcement, SSE `response.output_text.delta` stream, and
+    // `credits_consumed` actual-cost capture. (grok-4.5 was deferred 2026-07-13
+    // because none of this was live; 4.6 is its activation.)
+    kieSlugOrModel: "grok-4-6",
+    vendor: "xai",
+    structuredOutputMode: "responses-json-schema",
+    supportsImages: true,
+    maxOutputTokens: 16384,
+    // KIE's documented enum, each level live-verified (echoed back) 2026-08-18.
+    // No `none`: the endpoint reasons unconditionally (see thinkingDefaultOn).
+    reasoningEfforts: ["low", "medium", "high", "xhigh"],
+    // Live-probed 2026-08-18: `temperature` is silently IGNORED (request echo
+    // stays at the 0.7 default), so never send it — same treatment as GPT-5.5+.
+    supportsTemperature: false,
+    // Reasons with NO reasoning param sent (effort defaults to "low" server-side
+    // — a trivial probe spent 169 of 170 output tokens on reasoning), so every
+    // call needs output headroom, not just xhigh.
+    thinkingDefaultOn: true
+  },
   {
     id: "claude-sonnet-5",
     displayName: "Claude Sonnet 5",
@@ -5056,6 +5108,25 @@ var LLM_MODEL_IDS = LLM_MODELS.map((m) => m.id);
 var STRUCTURED_VISION_MODELS = LLM_MODELS.filter(
   (m) => m.supportsImages && m.structuredOutputMode != null
 );
+var LLM_VENDOR_ORDER = ["anthropic", "google", "openai", "xai"];
+var LLM_VENDOR_LABELS = {
+  anthropic: "Anthropic",
+  google: "Google",
+  openai: "OpenAI",
+  xai: "xAI"
+};
+var TIER_RANK = { economy: 0, standard: 1, premium: 2 };
+function groupLlmModelsByVendor(models = LLM_MODELS) {
+  const groups = [];
+  for (const vendor of LLM_VENDOR_ORDER) {
+    const members = models.filter((m) => m.vendor === vendor).sort((a, b) => TIER_RANK[a.tier] - TIER_RANK[b.tier]);
+    if (members.length > 0) groups.push({ vendor, label: LLM_VENDOR_LABELS[vendor], models: members });
+  }
+  return groups;
+}
+function orderedLlmModels(models = LLM_MODELS) {
+  return groupLlmModelsByVendor(models).flatMap((g) => g.models);
+}
 function motionGraphicsFeature(engine) {
   return engine === "lottie" ? "motion-graphics-lottie" : "motion-graphics";
 }
@@ -5080,6 +5151,12 @@ var LLM_FEATURE_DEFAULTS = {
 var LLM_MODALITY_CAPS = {
   "gemini-3-flash": { image: true, video: true, audio: true },
   "gemini-3.6-flash": { image: true, video: true, audio: true },
+  // gemini-3.7-flash is IMAGE-ONLY by DECISION, not omission: full video+audio
+  // caps would auto-enroll it in VIDEO_ANALYSIS_LLM_MODELS (derived below) and
+  // force a video-analysis tier + pricing decision that is deliberately
+  // deferred while the smart-family A/B routes this model internally (#747).
+  // Flip these two flags ONLY together with that VA-side decision.
+  "gemini-3.7-flash": { image: true, video: false, audio: false },
   "gemini-3.1-pro": { image: true, video: true, audio: true },
   "claude-haiku-4.5": { image: true, video: false, audio: false },
   "claude-sonnet-4.6": { image: true, video: false, audio: false },
@@ -5090,6 +5167,7 @@ var LLM_MODALITY_CAPS = {
   "gpt-5.6-luna": { image: true, video: false, audio: false },
   "gpt-5.6-terra": { image: true, video: false, audio: false },
   "gpt-5.6-sol": { image: true, video: false, audio: false },
+  "grok-4.6": { image: true, video: false, audio: false },
   "claude-sonnet-5": { image: true, video: false, audio: false },
   "claude-opus-4.8": { image: true, video: false, audio: false },
   "claude-opus-5": { image: true, video: false, audio: false },
@@ -13478,6 +13556,8 @@ exports.LLM_MODEL_IDS = LLM_MODEL_IDS;
 exports.LLM_REASONING_EFFORTS = LLM_REASONING_EFFORTS;
 exports.LLM_ROUTE_DEFAULTS = LLM_ROUTE_DEFAULTS;
 exports.LLM_TEXT_INPUT_MAX = LLM_TEXT_INPUT_MAX;
+exports.LLM_VENDOR_LABELS = LLM_VENDOR_LABELS;
+exports.LLM_VENDOR_ORDER = LLM_VENDOR_ORDER;
 exports.LOCATION_ASSET_TYPES = LOCATION_ASSET_TYPES;
 exports.LOCATION_ASSET_VARIANTS = LOCATION_ASSET_VARIANTS;
 exports.LOCATION_ATMOSPHERE_PROVIDERS = LOCATION_ATMOSPHERE_PROVIDERS;
@@ -13875,6 +13955,7 @@ exports.getWeapon = getWeapon;
 exports.getWeaponLabel = getWeaponLabel;
 exports.groupByFamily = groupByFamily;
 exports.groupHandleId = groupHandleId;
+exports.groupLlmModelsByVendor = groupLlmModelsByVendor;
 exports.hasContiguousSegmentDurations = hasContiguousSegmentDurations;
 exports.hasFeature = hasFeature;
 exports.hexToRgbaArray = hexToRgbaArray;
@@ -13933,6 +14014,7 @@ exports.normalizeModelInput = normalizeModelInput;
 exports.normalizeNodeModelParams = normalizeNodeModelParams;
 exports.normalizePinterestUrl = normalizePinterestUrl;
 exports.normalizeRoleSlug = normalizeRoleSlug;
+exports.orderedLlmModels = orderedLlmModels;
 exports.parseAttributedDialogue = parseAttributedDialogue;
 exports.parseCharacterMentionToken = parseCharacterMentionToken;
 exports.parseGroupHandle = parseGroupHandle;