@hiper2d/ai-agents 0.40.0 → 0.41.0

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
@@ -848,6 +848,7 @@ var LLM_CONSTANTS = {
848
848
  GPT: "gpt",
849
849
  GPT_MINI: "gpt-mini",
850
850
  // Gemini 4 Argon sits above Pro; version-free id so a later Argon repoint keeps persisted ids.
851
+ GEMINI_NEON: "gemini-neon",
851
852
  GEMINI_ARGON: "gemini-argon",
852
853
  GEMINI_PRO: "gemini-pro",
853
854
  GEMINI_FLASH: "gemini-flash",
@@ -1034,6 +1035,21 @@ var SupportedAiModels = {
1034
1035
  // werewolf day-2 vote fixture took 16-34s (380-415 output tokens). At medium the same
1035
1036
  // prompt took 4-11s (75-110 output tokens) with the same vote target and a reason just
1036
1037
  // as specific — the same trade Grok 4.7 made on 2026-10-06.
1038
+ // Gemini 4 Neon (added 2026-11-18, id verified against GET /v1beta/models): 1M in / 64K
1039
+ // out. Same thinkingLevel dialect — accepts low|medium|high, rejects 'minimal' and 'none'
1040
+ // (probed). Pinned LOW: it thinks even there (200-1,500 tokens a turn where Argon at medium
1041
+ // reports none). Werewolf day-3 prompt (10.5K tokens, cold cache, 2 runs, medians):
1042
+ // low = discussion 8.9s, vote 12.9s, night 10.7s, end-of-day notes 36s, 3.8-5.9¢ a turn;
1043
+ // medium = 15-16s on the short turns for no visible gain; high = 21-23s, notes 51s.
1044
+ // No `temperature` (see the Pro note). Speed tag owed until it has 100 production turns.
1045
+ [LLM_CONSTANTS.GEMINI_NEON]: {
1046
+ displayName: "Gemini 4 Neon",
1047
+ modelApiName: "gemini-4-neon",
1048
+ apiKeyName: API_KEY_CONSTANTS.GOOGLE,
1049
+ hasThinking: true,
1050
+ reasoningEffort: "low",
1051
+ tags: ["expensive"]
1052
+ },
1037
1053
  [LLM_CONSTANTS.GEMINI_ARGON]: {
1038
1054
  displayName: "Gemini 4 Argon",
1039
1055
  modelApiName: "gemini-4-argon",
@@ -1579,6 +1595,14 @@ var MODEL_PRICING = {
1579
1595
  extendedContextThresholdTokens: 1e5
1580
1596
  },
1581
1597
  // Google models
1598
+ [SupportedAiModels[LLM_CONSTANTS.GEMINI_NEON].modelApiName]: {
1599
+ // ai.google.dev pricing page, read 2026-11-18: one flat tier over the whole 1M window
1600
+ // (no long-context rows). Implicit caching only, cached input at 5% of input. Not on
1601
+ // Google's free tier.
1602
+ inputPrice: 3,
1603
+ outputPrice: 15,
1604
+ cacheHitPrice: 0.15
1605
+ },
1582
1606
  [SupportedAiModels[LLM_CONSTANTS.GEMINI_ARGON].modelApiName]: {
1583
1607
  // Launch pricing (ai.google.dev pricing page, fetched 2026-10-13), one flat tier — no
1584
1608
  // >200K long-context rows, unlike 3.1 Pro. DOUBLES on 2026-12-01 to $4.00 / $20.00 /
@@ -5667,6 +5691,7 @@ var AgentFactory = class {
5667
5691
  case LLM_CONSTANTS.GPT:
5668
5692
  case LLM_CONSTANTS.GPT_MINI:
5669
5693
  return new Gpt5Agent(name, instruction, model.modelApiName, key, model.temperature, shouldEnableThinking);
5694
+ case LLM_CONSTANTS.GEMINI_NEON:
5670
5695
  case LLM_CONSTANTS.GEMINI_ARGON:
5671
5696
  case LLM_CONSTANTS.GEMINI_PRO:
5672
5697
  case LLM_CONSTANTS.GEMINI_FLASH: