ARLoop releases fully-open looped LLMs: 1.2B beats Huginn-3.5B and Qwen3.5-2B
burny_tech · x · 2026-11-10
ARLoop is a family of fully-open pretrained looped transformers with Attention Residuals. The strongest model, ARLoop-1.2B, outperforms both Huginn-3.5B (a prior fully-open looped LLM) and Qwen3.5-2B Base on benchmarks across commonsense reasoning, math, and coding.
Noting that open weights are emerging but training recipes remain scarce, the team releases the full recipe: data preparation, training, intermediate checkpoints, and evaluation code.
Related event: ARLoop: Fully Open-Source Looped Transformers, 1.2B Beats Huginn-3.5B(5 posts)→
More from Research
- CoRL 2021 local chair recalls three-table demo area as show grows into CoRL 2026 — Ed__Johns · 2026-11-12
- Pretrained atomic potential screens 1,620 combos to stabilize one-atom-thick metals — bravo_abad · 2026-11-12
- Claude robot wins don't kill world models: why Fable 5.2 and World Labs' Atlas solve different problems — dotey · 2026-11-12
- Claude produced machine-checked proofs for 239 fuzz entry points across 123 libraries — kieranklaassen · 2026-11-12
- SynthICL: synthetic-data-only training hits 79% success on 16 real robot tasks — Ed__Johns · 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