@hiper2d/ai-agents 0.34.0 → 0.35.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/index.mjs CHANGED
@@ -1046,14 +1046,27 @@ var SupportedAiModels = {
1046
1046
  tags: ["expensive"]
1047
1047
  },
1048
1048
  [LLM_CONSTANTS.GEMINI_FLASH]: {
1049
- // Repointed 3.6 → 3.7 (2026-08-13) → 3.8 (2026-09-02); stable picker id, same pattern as gpt.
1050
- // 3.7 rejected thinkingLevel 'minimal' (low|medium|high only), unlike 3.5/3.6 — 3.8 untested
1051
- // on 'minimal', so keep the pin at 'medium' or above.
1052
- displayName: "Gemini 3.8 Flash",
1053
- modelApiName: "gemini-3.8-flash",
1049
+ // Repointed 3.6 → 3.7 (2026-08-13) → 3.8 (2026-09-02) → Gemini 4 Hydrogen (2026-11-12);
1050
+ // stable picker id, same pattern as gpt — the slot keeps the name `gemini-flash` although
1051
+ // Google dropped "Flash" from the model's name. Id verified against GET /v1beta/models
1052
+ // (1M in / 65K out). Accepts low|medium|high, rejects 'minimal' (400, probed — as did
1053
+ // 3.7 and 3.8 Flash).
1054
+ // Pinned 'low' (3.8 Flash ran 'medium'): Hydrogen is twice the price and thinks about
1055
+ // twice as much per level. Werewolf day-2 vote fixture, 3 runs a level, 2026-11-12:
1056
+ // Hydrogen low 3.4-3.9s 211-263 thinking tokens 0.58-0.60¢
1057
+ // Hydrogen medium 5.4-6.7s 528-761 0.81-0.98¢
1058
+ // Hydrogen high 6.4-7.8s 839-1175 1.03-1.29¢
1059
+ // 3.8 Flash medium 5.1-7.1s 203-315 0.29-0.34¢
1060
+ // 3.8 Flash low 2.1-2.3s 0 0.21¢
1061
+ // So 'low' is the level that thinks as much as the model it replaces did, a little
1062
+ // faster, at twice the cost instead of three times; every run named the same vote target
1063
+ // with a reason as specific. No `temperature` (see the Pro note). No speed tag until it
1064
+ // has 100+ production turns (3.8 Flash had none: p50 9.9s).
1065
+ displayName: "Gemini 4 Hydrogen",
1066
+ modelApiName: "gemini-4-hydrogen",
1054
1067
  apiKeyName: API_KEY_CONSTANTS.GOOGLE,
1055
1068
  hasThinking: true,
1056
- reasoningEffort: "medium"
1069
+ reasoningEffort: "low"
1057
1070
  },
1058
1071
  [LLM_CONSTANTS.GEMINI_LITE]: {
1059
1072
  displayName: "Gemini 3.5 Flash Lite",
@@ -1508,10 +1521,11 @@ var MODEL_PRICING = {
1508
1521
  [SupportedAiModels[LLM_CONSTANTS.CLAUDE_FABLE].modelApiName]: {
1509
1522
  // Full 1M context window at standard pricing (no extended-context premium).
1510
1523
  // Fable 5.1 (2026-09-02): same $10/$50 as Fable 5, but cache reads dropped to $0.25/MTok.
1511
- // Fable 5.2 (2026-11-10): same prices as 5.1.
1512
- inputPrice: 10,
1513
- outputPrice: 50,
1514
- cacheHitPrice: 0.25
1524
+ // Fable 5.2 (2026-11-10): $8/$40, cache reads $0.20 (pricing page, 2026-11-11) — 20% under
1525
+ // 5.1's $10/$50/$0.25. Cache writes $10 (5m) / $16 (1h) = the usual 1.25x / 2x of input.
1526
+ inputPrice: 8,
1527
+ outputPrice: 40,
1528
+ cacheHitPrice: 0.2
1515
1529
  },
1516
1530
  [SupportedAiModels[LLM_CONSTANTS.CLAUDE_OPUS].modelApiName]: {
1517
1531
  // Opus 5.5 (2026-09-22): $4/$20, down from Opus 5's $5/$25. Cache reads are 5% of input
@@ -1560,13 +1574,12 @@ var MODEL_PRICING = {
1560
1574
  extendedContextThresholdTokens: 2e5
1561
1575
  },
1562
1576
  [SupportedAiModels[LLM_CONSTANTS.GEMINI_FLASH].modelApiName]: {
1563
- // 3.8 Flash (2026-09-02) launched at the same rates as 3.7. 3.7's launch pricing was
1564
- // scheduled to double to $1.50/$7.50/$0.15 on 2027-01-01 (ai.google.dev pricing page,
1565
- // fetched 2026-08-13) — ACTION NEEDED then: re-check whether 3.8 follows and update.
1566
- // Cache storage cost ($0.50 / 1M tokens per hour) is not tracked here — the
1567
- // schema only models per-token call costs, not time-based storage.
1568
- inputPrice: 0.75,
1569
- outputPrice: 3.75,
1577
+ // Gemini 4 Hydrogen (ai.google.dev pricing page, Standard tier, read 2026-11-12): one
1578
+ // flat tier, twice the input and output rates of the 3.8 Flash it replaced
1579
+ // ($0.75 / $3.75) with the same $0.075 cached rate (5% of input, as on Argon). Implicit
1580
+ // caching only — explicit caching is not supported, so there is no storage cost.
1581
+ inputPrice: 1.5,
1582
+ outputPrice: 7.5,
1570
1583
  cacheHitPrice: 0.075
1571
1584
  },
1572
1585
  [SupportedAiModels[LLM_CONSTANTS.GEMINI_LITE].modelApiName]: {