Dev shows Hermes delegation routes: cost-weighted routing across six models per task type
intellectronica · x · 2026-10-30
A developer shares his "Hermes delegation routes" setup: requests are routed by task type and cost weight — default to deepseek-v4.1-flash (cost 0.2), smart coding/analysis to gpt-6.1-sol (0.7), fast tasks to gpt-6-luna (0.1), writing to muse-spark-1.3-contributor, agentic tool-heavy missions to gpt-6.1-terra, and hardest planning/review to gpt-6-astra (0.9). A clean example of tiered multi-model orchestration on one subscription.
More from coding & agent
- Cisco bets on 2-person team with unlimited AI tokens to rewrite 98% of legacy code in 24 months — dosco · 2026-10-30
- EdgeQuake v0.38: a full GraphRAG stack in Rust with 2.3k GitHub stars — JeremyCMorgan · 2026-10-30
- Rules MCP Server gives coding agents on-demand contextual rules and patterns — modelcontextprotocol · 2026-10-30
- load-lock: an open-source machine-wide semaphore that makes parallel coding agents take turns — erdar2 · 2026-10-30
- Claude Code config reveals block on recreating existing video games, sparking developer backlash — mark_k · 2026-10-30
- Turning existing Next.js API procedures into MCP tools: one public server, one admin — piotrkulpinski · 2026-10-30