diegosouzapw e063eae727 feat(clawrouter): implement 14 ClawRouter-inspired features ...

PRICING UPDATES (01-09):
- xAI Grok-4 family: grok-4-fast-non-reasoning (/usr/bin/bash.20/$0.50/M, 1143ms),
  grok-4-fast-reasoning, grok-4-1-fast-*, grok-4-0709, grok-3, grok-3-mini
- Z.AI GLM-5 family: glm-5 + glm-5-turbo (128k maxOutput, $1.00/$3.20/M)
- Gemini Flash Lite: price corrected $0.15→$0.10 / $1.25→$0.40 (per ClawRouter)
- Gemini 3.1 Pro: new flagship (1.05M context, aliased as gemini-3.1-pro)
- Anthropic Claude 4.5/4.6: haiku-4.5 ($1/$5), sonnet-4.6 ($3/$15), opus-4.6 ($5/$25)
- DeepSeek native section: deepseek-chat/v3/v3.2 ($0.28/$0.42), deepseek-reasoner ($0.55/$2.19)
- Kimi K2.5 Moonshot: kimi-k2.5 ($0.60/$3.00, 262k ctx), moonshot-kimi-k2.5 alias
- MiniMax M2.5: minimax-m2.5 ($0.30/$1.20, 204k ctx, reasoning+tools)
- NVIDIA free tier: gpt-oss-120b at $0.00/M via emergencyFallback.ts

INFRASTRUCTURE FEATURES (10-14):
- feat(router): add intentClassifier.ts for multilingual intent detection (9 langs)
  Detects code/reasoning/simple in EN, PT-BR, ES, ZH, JA, RU, DE, KO, AR
- feat(dedup): add requestDedup.ts for concurrent request deduplication
  SHA-256 hash, skip streaming, skip high-temperature, 60s failsafe TTL
- feat(autoCombo): add routerStrategy.ts pluggable strategy system
  RouterStrategy interface, RulesStrategy (6-factor) + CostStrategy, registry
- feat(fallback): add emergencyFallback.ts budget-exhaustion detector
  Triggers on HTTP 402 or budget keywords, redirects to nvidia/gpt-oss-120b
- feat(taskFitness): add fitness scores for Grok-4, Kimi K2.5, GLM-5,
  MiniMax M2.5, DeepSeek V3.2, Gemini 3.1 Pro across all task categories

PROVIDERS:
- providers.ts: add Z.AI (zai) provider entry for GLM-5 API key connections

All features on branch: feat/clawrouter-improvements
Source: github.com/BlockRunAI/ClawRouter analysis (2026-03-17)

2026-03-17 10:43:12 -03:00

6.8 KiB

Raw Blame History

View Raw