Part ⅩⅢ · Signal monitor
Follow the benchmark as it moves.
One place to track ARC-AGI — the official dispatches, a live @arcprize feed, topic-wide searches, the people and labs building it, and the newest ARC-related arXiv papers. Updated July 28, 2026.
Official dispatches
Latest from ARC Prize.
The most recent posts from the ARC Prize blogARC Prize uses its blog to announce major benchmark results, including the December 2024 o3 result: 75.7% at high efficiency and 87.5% at roughly 172× the compute.. The newest appears first; each card opens the original dispatch.
ARC-AGI-3 Milestone Prize #1
Tufa Labs' The Duck took first place, followed by Reki and forge. The official recap links each open submission and explains what worked.
Read dispatch ↗Analyzing GPT-5.5 & Opus 4.7 with ARC-AGI-3
ARC Prize analyzes 160 public replays and reasoning traces to identify three recurring world-model and transfer failures.
Read dispatch ↗Measuring human performance on ARC-AGI-3
ARC Prize released its controlled study of 458 participants, the human action-efficiency baselines, and the resulting scoring updates.
Read dispatch ↗Live social feed
Twitter / X · @arcprize
Loading the live @arcprize feed…
Research archive
Recent ARC-related arXiv papers.
Newest first: direct ARC-AGI work plus related human and cognitive-science validationHuman studies establish the reference point. On ARC-AGI-2, people solve essentially 100% of evaluation tasks while the best 2025 machine reached about 24%.. Each row opens the arXiv abstract.
- Bringing Back Rule Induction to Fluid Intelligence Research? An Initial Validation of the ARC-AGI Benchmark in Humans Human validation
- Structural Grid Descriptors Predict Within-Task Solver Success on ARC-AGI Symbolic solvers
- Multi-Perspective Transformers in ARC-AGI-2 Challenge Test-time training
- ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence Benchmark
- Exploring Human Behavior During Abstract Rule Inference and Problem Solving with the Cognitive Abstraction and Reasoning Corpus Cognitive science
- ARC Prize 2025: Technical Report Competition report
Model frontier
Labs reporting ARC results
The ARC Prize 2025 technical report names these four frontier labs as public ARC-AGI reporters. Treat model-card posts as claims until ARC Prize marks a score verified.
- Anthropic @AnthropicAI
- Google DeepMind @GoogleDeepMind
- OpenAI @OpenAI
- xAI @xai
Long-form signal
Blogs, Substacks & lab notes
Next official checkpoint
ARC-AGI-3 Milestone Prize #2 · September 30, 2026
Final competition submissions follow on November 2; paper submissions close November 8.