New Paper: 150M Model Achieves 29.5% on ARC-AGI at $0.0007 per Task
burny_tech · x · 2026-08-15
The paper 'BDH-CQ: In-Context Learning with Recurrent Latent Reasoning' proposes a method that updates a recurrent memory with examples and reasons in a continuous latent state without verbalizing intermediate steps. A 150M model reaches 29.5% on ARC-AGI-1 at a cost of $0.0007 per task, setting a new cost-efficiency frontier.
More from Research
- UniReps 4th edition satellite event at NeurIPS 2026 in Paris, seeking input — ClementineDomi6 · 2026-08-15
- PlasmidGPT Designs DNA Sequences Trained on 153k Plasmids — anshulkundaje · 2026-08-15
- ProteinDPO Adopts RLHF Technique to Train More Stable Protein Sequences — BrianHie · 2026-08-15
- Running H3 video generation on 4GB VRAM: a 3050 laptop success story — yushairiegalaxy96 · 2026-08-15
- View: Task Completion is Low-Dimensional; RL Transfers "Dark Knowledge" — kalomaze · 2026-08-15
- Model Cascade Achieves 90% Cost Savings: Shreya Shares Classification Optimization Tips — HamelHusain · 2026-08-15