Teknium doubts Jev can deliver meaningful context compaction or cheap-model routing
Teknium · x · 2026-09-19
After a user asked Teknium to update Hermes for Jev — or build a mini harness using Jev to route prompts to the cheapest capable LLM — he replied that Jev would in no way help with compaction that much. Part of the same exchange where he argues Hermes already summarizes 95% of context out.
Related event: Teknium doubts Jev's compression: Hermes summarizes 95% of context(2 posts)→
More from coding & agent
- Jev ecosystem explodes overnight: browser-use's jev-ultrafast hits ~8k stars, books flights in 7.1s — udmrzn · 2026-09-20
- Dev's custom reality-check skill returns a clean bill of health for beads_viewer — doodlestein · 2026-09-20
- TurnPanel launches public beta: a local-first AI workspace that operates your computer — Scobleizer · 2026-09-20
- Sutro Open-Sources jev-align, a CLI That Aligns Jev AI Functions With Your Judgment via GEPA — RichmanRonald · 2026-09-20
- Two vLLM bugs hid in plain sight: Mamba state cache, decode-before-prefill and a 32-bit wrap — AI Engineer · 2026-09-20
- Browser Use CEO: "browsers are dead" — CLI agents beat browser UIs 10x — annetgriffin · 2026-09-20