150M Recurrent Model Scores 29.5% on ARC-AGI at $0.0007 per Task
juanviera23 · reddit · 2026-08-15
A paper from the Pathway team presents a 150M parameter recurrent model achieving 29.5% on ARC-AGI-1 at a cost of $0.0007 per task. Unlike Transformers, it uses a recurrent latent reasoning setup that "thinks" in latent space before answering. This result sits completely outside the published cost/accuracy frontier for ARC-AGI, and the model's small size allows it to run on virtually any hardware. The author is eager to see it scaled up to 1-3B parameters.
Related event: 150M Model Achieves 29.5% on ARC-AGI(3 posts)→
More from Research
- Researcher slams papers whose stacked changes beat even weak baselines barely — retr0jirachi · 2026-10-16
- FrankenSim builds a physics-based virtual orchestra, piano and drum demos live — doodlestein · 2026-10-16
- FoveaVLA replaces wrist cameras with learned foveation, hitting 70.8% success — Haoyu_Xiong_ · 2026-10-16
- ApolloPFN: a zero-shot forecasting network that natively handles exogenous covariates — burkov · 2026-10-16
- ICLR reviewer says 4/5 papers in review pile read like AI slop — berkustun · 2026-10-16
- Tsinghua's DBS learns content-adaptive frequency band splits for audio source separation — thu-ml · 2026-10-16