What's New in v2.7.0
🔀 Inspired Routing Features
- Pluggable RouterStrategy — rules, cost, and latency strategies with
registerStrategy() - Multilingual intent detection — 30+ language keyword scoring for task routing
- Benchmark-driven fallback chains — auto-generated from performance signals
- Request deduplication — content-hash-based dedup with configurable time window
- toolCalling flag in provider registry for tool-aware routing
- modelCapabilities service for model introspection
💰 New Pricing & Providers
- Grok-4 Fast (xAI) —
grok-4-fast-non-reasoning/grok-4-fast-reasoning - GLM-5 / GLM-5-Turbo (Z.AI)
- MiniMax M2.5
- Kimi K2.5 + Kimi K2.5 Thinking
- Z.AI provider (anthropic-compatible endpoint)
🔧 Infrastructure
- MCP Server: 5 new tool schemas, 3 new auth scopes
- Health API: latency p50/p95 metrics per provider
- Observability: request tracking improvements
- SQLite WAL optimizations
🐛 Bug Fixes
- Fixed ambiguous model resolution for
claude-haiku-4-5-20251001(duplicate in providers) - Fixed Next.js App Router import resolution for
routerStrategy.ts - Excluded
app/vscode-extension/andapp/electron/from npm package