ARC Prize 2025: Technical Report

TL;DR

This paper surveys ARC Prize 2025 and its ARC-AGI-2 benchmark, highlighting how refinement loops—ranging from evolutionary program synthesis to zero-pretraining weight-space optimization—drive progress in abstract reasoning. It documents large-scale participation, with a top private-score of $24\%$, and showcases open-source winning approaches like the Tiny Recursive Model (7M parameters) and CompressARC (76K parameters) that exploit per-task refinement and MDL-based search to solve tasks without extensive pretraining. The analysis discusses knowledge-dependent limitations, benchmark contamination risks, and the need for continual benchmark adaptation (ARC-AGI-3) to probe interactive reasoning aspects such as exploration, planning, memory, and alignment. Overall, ARC Prize 2025 demonstrates open, iterative progress toward AGI, emphasizes refinement-loop physics as a core mechanism, and argues for evolving benchmarks to sustain meaningful progress while balancing knowledge coverage with verifiable feedback.

Abstract

The ARC-AGI benchmark series serves as a critical measure of few-shot generalization on novel tasks, a core aspect of intelligence. The ARC Prize 2025 global competition targeted the newly released ARC-AGI-2 dataset, which features greater task complexity compared to its predecessor. The Kaggle competition attracted 1,455 teams and 15,154 entries, with the top score reaching 24% on the ARC-AGI-2 private evaluation set. Paper submissions nearly doubled year-over-year to 90 entries, reflecting the growing research interest in fluid intelligence and abstract reasoning. The defining theme of 2025 is the emergence of the refinement loop -- a per-task iterative program optimization loop guided by a feedback signal. Refinement loops come in a variety of forms, in particular evolutionary program synthesis approaches and application-layer refinements to commercial AI systems. Such refinement loops are also possible in weight space, as evidenced by zero-pretraining deep learning methods which are now achieving competitive performance with remarkably small networks (7M parameters). In parallel, four frontier AI labs (Anthropic, Google DeepMind, OpenAI, and xAI) reported ARC-AGI performance in public model cards in 2025, establishing ARC-AGI as an industry standard benchmark for AI reasoning. However, our analysis indicates that current frontier AI reasoning performance remains fundamentally constrained to knowledge coverage, giving rise to new forms of benchmark contamination. In this paper, we survey the top-performing methods, examine the role of refinement loops in AGI progress, discuss knowledge-dependent overfitting, and preview ARC-AGI-3, which introduces interactive reasoning challenges that require exploration, planning, memory, goal acquisition, and alignment capabilities.

Figures (3)

• Figure 1: Example ARC-AGI task
• Figure 2: TRM architecture.
• Figure 3: CompressARC architecture and approach.