Context Language Models: Letting LLMs Edit Their Own Context Boosts Accuracy 11.4% While Cutting FLOPs 21.5%
datawithsuman · x · 2026-10-05
A paper by Rulin Shao, Luke Zettlemoyer, Pang Wei Koh et al. (AI2/UW) introduces Context Language Models (CLMs): models that natively manage their own context by treating it as a file they can freely edit.
Key results:
- Zero-shot CLMs beat SOTA context management strategies: +11.4% accuracy with 21.5% fewer FLOPs on BrowseComp-Plus, +5% with 59% fewer FLOPs on 12-hour EdgeBench, and 65% greater improvement at equal compute on a 24-hour multi-repo agent-swarm task.
- Moving context management from external harness into the model enables in-context and parametric learning of strategies; natural-language skill-evolution loops improve held-out accuracy by up to 35.9 points.
- An online RL method lifts Qwen3.5-9B by 47.6% on BrowseComp-Plus with 12% fewer FLOPs.
- Suffix Cache Reuse serving cuts server-side compute 35% vs standard SGLang.
Caveats: with a 128K budget the gap to summarization mostly closes (47.3 vs 47.8 on EdgeBench-10, though cost stays lower); some math results are single runs; editable context creates a new prompt-injection surface, as the authors flag.
More from coding & agent
- New TLAPS-Bench blog: formally proving correctness of real-world system specs with TLA+ — tianyin_xu · 2026-11-19
- Pierre releases Words, a streaming Markdown parser 2x faster than md4x and 20x faster than Streamdown — andersonbcdefg · 2026-11-19
- Ex-Claude superfan: getting the best from Astra/Sol requires relearning how you prompt — mariofilhoml · 2026-11-19
- Rivet's lesson: JEV doesn't reason—split severity classification into tiny yes/no prompts — yawnxyz · 2026-11-19
- 16x Microsoft MVP demos a native PowerShell agent built on Microsoft's Intelligent Terminal — dfinke · 2026-11-19
- Musk: Treat Grok Bot like a really smart employee, says most users underuse it — elonmusk · 2026-11-19