With CliffCompaction, open-weight models beat closed ones in long-horizon sessions
Tim_Dettmers · x · 2026-09-23
Tim Dettmers observes that CliffCompaction creates a usefulness gap between open-weight models and closed models that hide their reasoning traces: for long-horizon sessions, open-weight models are far stronger, and some models simply "compact better" than others. This builds on his point that the tool works much better with full thinking traces — consistent with frontier labs, where OpenAI's math and ARC-AGI results largely stem from trace-aware autocompaction.
More from coding & agent
- User automates entire video clipping workflow with Nous Research's Hermes agent — Teknium · 2026-11-12
- As models improve, writing good evals gets harder and more important — amankhan · 2026-11-12
- Prime agent ships with day-one multi-agent communication, runs in Minecraft unmodified — sarahcat21 · 2026-11-12
- Autonomous agent edits just 1,024 weights to lift Qwen3.5 0.8B from 7.4% to 37.4% on MATH-500 — VictorSanh · 2026-11-12
- AI game jam with Anthropic, Google, OpenAI teams: cash prizes plus $50K continuation grants — chrisfirst · 2026-11-12
- Free AI game jam in SF Nov 14–15 with teams from Anthropic, Google, Meshy, OpenAI — chrisfirst · 2026-11-12