Jev for Recursive Language Models: half the cost, 6x faster at 1M tokens
dbreunig · x · 2026-10-25
A deep-dive on Jev (System One), a sub-calling scheme for Recursive Language Models (RLMs). RLMs let a root LM work over contexts far beyond its window: the context lives in a REPL variable, the root writes code over it and calls sub-models on slices.
Key findings:
- half the cost and 6x faster at 1M tokens
- Experiments with abstention, routing to bigger models, and honest write-ups of where the approach didn't help
- A quoted follow-up: on OOLONG's TREC tasks, a DSPy ReAnchor-style method cut per-label count error from 7.9% to 2.6% and lifted the score from 0.75 to 0.81
More from coding & agent
- Hermes ships subagent routes plugin for per-task model orchestration — intellectronica · 2026-10-25
- Claude drives ComfyUI for hours, auto-generates ~500 varied Kromoss images — hazadaza · 2026-10-25
- How LLM-based penetration testing harnesses work: LLM plans, harness executes — utku1337 · 2026-10-25
- Dev builds a 12MB PvP mini-game that hooks into AI agents to stop idle scrolling — CallumReadIt · 2026-10-25
- Open-source parametric CAD with one interface for AIs and humans, parts by Claude — UnspeakableTruths · 2026-10-25
- EMNLP paper localizes faithfulness and citation errors in agentic deep research systems — EliasEskin · 2026-10-25