aiqbench analysis
This research introduces CHARM, a hybrid model that leverages structured task memory and synthetic data augmentation to significantly boost performance on ARC-style reasoning benchmarks. By employing compositional sparse embeddings, the approach reduces task-memory parameters by over 90% while maintaining high accuracy, offering a more efficient path toward scalable recurrent reasoning systems. The findings demonstrate that balancing recurrent depth with learning horizon is critical for optimizing model capacity without relying solely on massive backbones.
Structured Sparse Memory for Recurrent Reasoning
arXiv:2609.33270v1 Announce Type: new Abstract: Recurrent models trained from scratch have recently become competitive on ARC-style reasoning tasks, but the usual framing around small recurrent backbones overlooks two important parts of the system: task-conditioned memory and synthetic augmentation data. We study this regime through CHARM, a compact hybrid ARC model that combines recurrent reasoning with structured task memory, synthetic data, and inference-time aggregation. In existing approaches, task-conditioned memory supplies a large hidden source of capacity, reaching more than 30x the size of the recurrent backbone. We introduce a compositional sparse embedding (CoSE) for task conditioning that reduces learned task-memory parameters by over 90% while improving pass@2 in controlled ARC ablations. For the recurrent backbone, recurrent depth helps only when balanced with learning horizon. Combining these ingredients, our system reaches 84% pass@2 on ARC-AGI-1 and 46.7% pass@2 on ARC-AGI-2 public evaluation. The benefits of structured memory also generalize to unseen puzzles and other domains. Our code, dataset, and model checkpoints are available at https://github.com/water-vapor/charm.
Abstract:Recurrent models trained from scratch have recently become competitive on ARC-style reasoning tasks, but the usual framing around small recurrent backbones overlooks two important parts of the system: task-conditioned memory and synthetic augmentation data. We study this regime through CHARM, a compact hybrid ARC model that combines recurrent reasoning with structured task memory, synthetic data, and inference-time aggregation. In existing approaches, task-conditioned memory supplies a large hidden source of capacity, reaching more than 30x the size of the recurrent backbone. We introduce a compositional sparse embedding (CoSE) for task conditioning that reduces learned task-memory parameters by over 90% while improving pass@2 in controlled ARC ablations. For the recurrent backbone, recurrent depth helps only when balanced with learning horizon. Combining these ingredients, our system reaches 84% pass@2 on ARC-AGI-1 and 46.7% pass@2 on ARC-AGI-2 public evaluation. The benefits of structured memory also generalize to unseen puzzles and other domains. Our code, dataset, and model checkpoints are available at this https URL.
Artificial Intelligence (cs.AI)
arXiv:2609.33270 [cs.AI]
(or arXiv:2609.33270v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.33270 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 254 words
Related coverage
Pathway’s brain-inspired architecture development on Amazon SageMaker HyperPod
AWS · Sep 8, 2026
[AINews] Claude Fable/Mythos 5.1: new SOTA model, 75% cache price cut but 70% more output tokens
Latent Space · Sep 2, 2026
Deploying Qwen3.8-2.4T-A95B on Amazon SageMaker HyperPod with vLLM
AWS · Sep 9, 2026
SANE: State Anomaly Neutralization for Stable Extreme-Context Delta-Rule Models
arXiv · Aug 25, 2026