7M-Parameter Tiny Recursion Model Hits 45% on ARC-AGI-1
CatAstro_Piyush · x · 2026-09-04
The paper 'Less is More: Recursive Reasoning with Tiny Networks' introduces TRM, a 7M-parameter network that recursively refines its answer over up to K steps, scoring 45% on ARC-AGI-1 and 8% on ARC-AGI-2. It simplifies the Hierarchical Reasoning Model by dropping biological arguments, fixed-point theorems, and hierarchy. Small enough to pretrain and analyze on a single GPU, TRM argues against the assumption that only massive foundation models can reason.
More from Models
- Suno now caps downloads: $20 tier gets only 20 audio files per month — aziz4ai · 2026-09-04
- GPT-6 Astra hits 66% on ARC-AGI-3, up from 8%, as ARC Prize plans AGI-4 around open-ended invention — GaryMarcus · 2026-09-04
- Scaling Law for Looped Transformers: Looping Boosts Reasoning, Not Knowledge — bookwormengr · 2026-09-04
- AI experts once pegged AGI at 2075-2100 — OpenAI's Astra shows how wrong they were — dee_hw · 2026-09-04
- Models still can't see determinants, but no-thinking time-horizons double every ~9 months — scaling01 · 2026-09-04
- GPT-6 Astra Ultra builds a Minecraft wooden house zero-shot in ~9 minutes, stairs perfect — Angaisb_ · 2026-09-04