Looped transformer lifts no-CoT time horizon from 16s to 37s, biggest gains on multi-hop reasoning
nrehiew_ · x · 2026-11-21
nrehiew reviews results from the newly released variable-looping research model TML-d24-varloop-research:
- More looping raises no-CoT capability with diminishing returns: the no-CoT 50% time horizon grows from 16s at one pass to 37s at three, then flattens; the model was never trained beyond five passes yet ten passes don't degrade it.
- Multi-hop reasoning benefits most: on N-Hop Lookup, more passes let the model chain more hops; with CoT, gains are largest on long-context retrieval, then math/science reasoning (HMMT, GPQA Diamond, HLE) and long-horizon agentic tasks (SWE-bench Pro, BrowseComp), while closed-book QA is nearly invariant.
- On the concern that looped transformers worsen CoT monitorability, the author credits the team for releasing models with varying loop counts for open research.
More from Models
- Gemini 4 Neon locked to Pro and Ultra paid subscribers, leaving Plus users out — giffmana · 2026-11-21
- Looped transformers: more looping boosts reasoning but with diminishing returns past 3 passes and fewer CoT tokens — nrehiew_ · 2026-11-21
- Team releases TML-d24-varloop-research on Hugging Face to study looping's effect on CoT monitorability — nrehiew_ · 2026-11-21
- Full sub-benchmark breakdown for Inkling 2 Mini published by Artificial Analysis — ArtificialAnlys · 2026-11-21
- Inkling 2 Mini costs ~6x less and runs ~4x faster per task than Qwen3.8 27B, AA finds — ArtificialAnlys · 2026-11-21
- Inkling 2 Mini matches 27B-33B rivals on intelligence with ~5-6x fewer active parameters — ArtificialAnlys · 2026-11-21