@genesislcap/ai-assistant 15.88.2 → 15.89.0

package/dist/chat-driver.mjs

@@ -1315,6 +1315,7 @@ var SUPPORTED_GEMINI_MODEL_IDS = [
   "gemini-2.5-pro",
   "gemini-2.5-flash",
   "gemini-2.5-flash-lite",
+  "gemini-4-argon",
   "gemini-3.8-flash",
   "gemini-3.5-flash",
   "gemini-3.1-flash-lite",
@@ -1403,6 +1404,7 @@ function anthropicTokenCost(model, usage) {
 var GEMINI_LONG_CONTEXT_THRESHOLD = 2e5;
 var GEMINI_CACHED_INPUT_MULTIPLIER = 0.1;
 var GEMINI_3_8_FLASH_INTRO_RATE_ENDS = Date.parse("2027-01-01T00:00:00Z");
+var GEMINI_4_ARGON_LAUNCH_RATE_ENDS = Date.parse("2026-12-01T00:00:00Z");
 var GEMINI_PRICING = {
   "gemini-2.5-flash-lite": {
     kind: "flat",
@@ -1419,37 +1421,50 @@ var GEMINI_PRICING = {
     rate: { promptPerMillion: 0.25, candidatePerMillion: 1.5 }
   },
   "gemini-3.5-flash": { kind: "flat", rate: { promptPerMillion: 1.5, candidatePerMillion: 9 } },
-  // INTRODUCTORY RATE, and the only dated entry in this file. Google publishes
-  // $0.75 / $3.75 per MTok **through 2026-12-31**, rising to $1.50 / $7.50 on 2027-01-01 —
-  // so today this model is CHEAPER than gemini-3.5-flash above and in January it is not.
+  // INTRODUCTORY RATE. Google publishes $0.75 / $3.75 per MTok **through 2026-12-31**, rising
+  // to $1.50 / $7.50 on 2027-01-01 — so until then this model is CHEAPER than gemini-3.5-flash
+  // above, and from then it is not.
   //
-  // The listed figure is the one in force, not the one coming. That is the lesson the
-  // `claude-sonnet-5` entry records from the other direction: charging a scheduled future
-  // rate early over-billed every request by 50% for as long as nobody noticed. A rate that
-  // has not taken effect is not a price, and `GeminiPricing` deliberately has no time
-  // dimension to express one — flip these two numbers on 2027-01-01 instead.
-  //
-  // That flip is ENFORCED, not remembered: `token-cost.test.ts` carries a dated assertion that
-  // turns the build red on 2027-01-01 and names the three-line edit (these two rates, plus the
-  // now-false `< gemini-3.5-flash` comparison in that same test — both models are $1.50 input
-  // from that date). There is no ticket to lose.
+  // `dated`, carrying BOTH rates and the switch instant, rather than today's rate plus a test
+  // that goes red on the day. Each request is priced at the rate in force when it is made, so a
+  // scheduled future rate is never charged early — the trap the `claude-sonnet-5` entry records,
+  // where that over-billed every request by 50% — and the switch needs no edit and no release.
+  // That second half is the point: a failing test only reaches our own CI, while every version
+  // already published would keep billing the old rate until a fix shipped and apps upgraded.
+  // https://ai.google.dev/gemini-api/docs/pricing
+  "gemini-3.8-flash": {
+    kind: "dated",
+    before: { promptPerMillion: 0.75, candidatePerMillion: 3.75 },
+    after: { promptPerMillion: 1.5, candidatePerMillion: 7.5 },
+    switchAtMs: GEMINI_3_8_FLASH_INTRO_RATE_ENDS
+  },
+  // Launch rate $2 / $10 per MTok until 30 Nov 2026, then $4 / $20 — `dated`, like
+  // gemini-3.8-flash above. Selected by the `at` argument of `geminiRatesFor` (default: now).
   // https://ai.google.dev/gemini-api/docs/pricing
-  "gemini-3.8-flash": { kind: "flat", rate: { promptPerMillion: 0.75, candidatePerMillion: 3.75 } },
+  "gemini-4-argon": {
+    kind: "dated",
+    before: { promptPerMillion: 2, candidatePerMillion: 10 },
+    after: { promptPerMillion: 4, candidatePerMillion: 20 },
+    switchAtMs: GEMINI_4_ARGON_LAUNCH_RATE_ENDS
+  },
   "gemini-3.1-pro-preview": {
     kind: "tiered",
     standard: { promptPerMillion: 2, candidatePerMillion: 12 },
     longContext: { promptPerMillion: 4, candidatePerMillion: 18 }
   }
 };
-function geminiRatesFor(model, promptTokens) {
+function geminiRatesFor(model, promptTokens, at = /* @__PURE__ */ new Date()) {
   const pricing = GEMINI_PRICING[model];
   if (pricing.kind === "flat") {
     return pricing.rate;
   }
+  if (pricing.kind === "dated") {
+    return at.getTime() < pricing.switchAtMs ? pricing.before : pricing.after;
+  }
   return promptTokens > GEMINI_LONG_CONTEXT_THRESHOLD ? pricing.longContext : pricing.standard;
 }
-function geminiTokenCost(model, usage) {
-  const { promptPerMillion, candidatePerMillion } = geminiRatesFor(model, usage.promptTokens);
+function geminiTokenCost(model, usage, at = /* @__PURE__ */ new Date()) {
+  const { promptPerMillion, candidatePerMillion } = geminiRatesFor(model, usage.promptTokens, at);
   const m = TOKENS_PER_MILLION;
   const uncachedPromptTokens = Math.max(0, usage.promptTokens - usage.cachedTokens);
   const promptCost = uncachedPromptTokens / m * promptPerMillion;
@@ -3043,6 +3058,7 @@ var GEMINI_CONTEXT_LIMITS = {
   "gemini-2.5-pro": 1048576,
   "gemini-2.5-flash": 1048576,
   "gemini-2.5-flash-lite": 1048576,
+  "gemini-4-argon": 1048576,
   "gemini-3.8-flash": 1048576,
   "gemini-3.5-flash": 1048576,
   "gemini-3.1-flash-lite": 1048576,
@@ -3068,10 +3084,15 @@ var GEMINI_MODEL_WARNINGS = {
   "gemini-2.5-flash": "GeminiTransport: using gemini-2.5-flash \u2014 higher cost than flash-lite; use for harder reasoning or agent tasks.",
   "gemini-2.5-pro": "GeminiTransport: using gemini-2.5-pro \u2014 significantly higher, prompt-size-tiered cost; reserve for tasks where flash reliability is insufficient.",
   "gemini-3.5-flash": "GeminiTransport: using gemini-3.5-flash \u2014 frontier model, materially higher cost than flash-lite; use for harder reasoning or agent tasks.",
-  // Warns despite currently costing LESS than gemini-3.5-flash, because that is temporary: the
-  // rate doubles on 2027-01-01 (see GEMINI_PRICING). A caller that picks this tier on today's
-  // price and never revisits the choice gets twice the bill in January without changing a line.
-  "gemini-3.8-flash": "GeminiTransport: using gemini-3.8-flash \u2014 frontier model, materially higher cost than flash-lite. Its $0.75/$3.75 per-MTok rate is introductory and DOUBLES to $1.50/$7.50 on 2027-01-01; re-check this tier choice before then.",
+  // Warns despite costing LESS than gemini-3.5-flash until 2027-01-01, because that is temporary:
+  // the rate doubles that day (see GEMINI_PRICING, which switches on its own). A caller that picks
+  // this tier on the introductory price and never revisits the choice gets twice the bill in
+  // January without changing a line. Worded to stay true after the switch.
+  "gemini-3.8-flash": "GeminiTransport: using gemini-3.8-flash \u2014 frontier model, materially higher cost than flash-lite. Billed at an introductory $0.75/$3.75 per MTok through 2026-12-31 and $1.50/$7.50 from 2027-01-01; weigh a tier choice against the later rate.",
+  // Like gemini-3.8-flash, warns about a scheduled rate change: $2/$10 per MTok until 2026-11-30,
+  // then $4/$20 (see GEMINI_PRICING, which switches on its own). Worded to stay true after the
+  // switch, so the message needs no edit on the day.
+  "gemini-4-argon": "GeminiTransport: using gemini-4-argon \u2014 premium reasoning model, materially higher cost than the flash tiers. Billed at $2/$10 per MTok through 2026-11-30 and $4/$20 from 2026-12-01; weigh a tier choice against the later rate.",
   "gemini-3.1-pro-preview": "GeminiTransport: using gemini-3.1-pro-preview \u2014 PREVIEW model (no stability guarantee, may be withdrawn) with significantly higher, prompt-size-tiered cost; reserve for tasks where flash reliability is insufficient."
 };
 var GEMINI_THINKING_DISABLEABLE = [
@@ -3084,8 +3105,18 @@ var GEMINI_MEDIA_RESOLUTION_TIERS = [
   "gemini-3.1-flash-lite",
   "gemini-3.1-pro-preview"
 ];
+var GEMINI_SCHEMA_WITH_TOOLS_TIERS = [
+  "gemini-4-argon",
+  "gemini-3.8-flash",
+  "gemini-3.5-flash",
+  "gemini-3.1-flash-lite",
+  "gemini-3.1-pro-preview"
+];
 var GEMINI_LEVEL_LADDER = ["minimal", "low", "medium", "high"];
 var GEMINI_THINKING_LEVELS = {
+  // Same accepted set as gemini-3.8-flash (Google docs, 2026-10-29: low/medium/high, default
+  // medium, no `minimal`) — documented rather than probed, so re-measure when convenient.
+  "gemini-4-argon": ["low", "medium", "high"],
   "gemini-3.8-flash": ["low", "medium", "high"],
   "gemini-3.5-flash": ["minimal", "low", "medium", "high"],
   "gemini-3.1-flash-lite": ["minimal", "low", "medium", "high"],
@@ -3173,8 +3204,8 @@ var GeminiTransport = class _GeminiTransport {
    *   default posture. Without this, a caller asking for `high` on `gemini-2.5-pro` and quietly
    *   getting the default is left to wonder why.
    * - A graded depth whose DIALECT a Gemini 3 tier speaks but whose particular value it does not
-   *   accept — `minimal` on `gemini-3.8-flash` or `gemini-3.1-pro-preview`, the two tiers that
-   *   reject it outright. Nothing is dropped here; `clampThinkingLevel` MOVES the level onto the
+   *   accept — `minimal` on `gemini-4-argon`, `gemini-3.8-flash` or `gemini-3.1-pro-preview`, the tiers
+   *   that reject it outright. Nothing is dropped here; `clampThinkingLevel` MOVES the level onto the
    *   nearest one the tier takes, and since ties resolve upward the move is usually to something
    *   DEARER than the caller chose. A cost decision reversed on the wire in the expensive
    *   direction is exactly what this warning exists for — without it the only trace is the bill.
@@ -3303,7 +3334,7 @@ var GeminiTransport = class _GeminiTransport {
         });
       }
       const offeredToolNames = new Set(((_c = options === null || options === void 0 ? void 0 : options.tools) !== null && _c !== void 0 ? _c : []).map((t) => t.name));
-      const schemaCoexistsWithTools = this.model.startsWith("gemini-3");
+      const schemaCoexistsWithTools = GEMINI_SCHEMA_WITH_TOOLS_TIERS.includes(this.model);
       const applyResponseSchema = (options === null || options === void 0 ? void 0 : options.responseSchema) != null && (!tools || schemaCoexistsWithTools);
       const response = yield this.post(Object.assign({
         model: this.model,
@@ -3816,7 +3847,7 @@ var DEFAULT_AI_TIERS = Object.freeze({
       timeoutMs: AI_NO_RESPONSE_TIMEOUT_MS
     }),
     // On an INTRODUCTORY rate that doubles to $1.50 / $7.50 on 2027-01-01, from which point it
-    // is dearer on input than the gemini-3.5-flash it replaced. Chosen at the introductory rate
+    // no longer undercuts the gemini-3.5-flash it replaced on input. Chosen at the introductory rate
     // knowingly; `ai-tiers.test.ts` fails that day to force the re-bench.
     high: Object.freeze({
       model: "gemini-3.8-flash",