Yutori ships Navigator 2-5x cheaper: 80% prefix cache hit rate, Qwen post-training
togethercompute · x · 2026-08-21
Web agent economics are driven by inference volume: a simple task (extract a field, fill a form, navigate) takes 10-20 inference calls, and multi-site workflows run into the hundreds. Yutori's co-founder/CEO explained at Raise Summit how Navigator runs 2-5x cheaper and faster than closed models on web tasks:
- 80% prefix cache hit rate on repetitive contexts
- Speculative decoding on smaller-footprint models
- Post-training Qwen models with SFT and RL instead of a from-scratch base
The team runs on Together, where customizable endpoints and auto-scaling let them A/B test new model variants in minutes.
More from coding & agent
- Open-source AIUsage: one dashboard for quotas, costs and accounts across 12+ AI subscriptions — tom_doerr · 2026-08-21
- Stanford researcher: automated orchestrators are a big unlock, current versions not there yet — anshulkundaje · 2026-08-21
- CopilotKit open-sources OpenBot: AI coworkers that each get their own computer — Roger_M_Taylor · 2026-08-21
- Ornith 1.5 35B Q5 runs locally inside GitHub Copilot on a Mac M3 Max — DanWahlin · 2026-08-21
- Developer ditches MacBook Pro, moves all agentic work to a Linux machine — Rasmic · 2026-08-21
- Building an AI-native email client on Cloudflare Email, Workers and AI Gateway — ritakozlov · 2026-08-21