ARC-AGI-3
Also written: ARC, AI2 Reasoning Challenge, ARC-AGI-2, ARC-AGI, ARC-AGI-3 benchmark, AI and Reasoning Challenge - Artificial General Intelligence, ARC-AGI-1
A benchmark for AI abstract reasoning and problem-solving, measuring progress toward artificial general intelligence.
ARC-AGI-3 is a test designed to see how well an AI can think and solve problems, especially those that require abstract reasoning. It's like a difficult puzzle or logic test for computers.
Imagine you have to figure out a pattern or rule based on a few examples, and then apply that rule to a new situation. ARC-AGI-3 presents AI with tasks like this, where it needs to understand underlying principles rather than just memorizing information. High scores on this test suggest an AI is getting closer to human-like general intelligence.
The AI2 Reasoning Challenge (ARC) is a dataset and benchmark designed to evaluate an AI's abstract reasoning capabilities. It consists of multiple-choice questions that require understanding and applying concepts across diverse domains, moving beyond pattern matching to assess deeper problem-solving skills. The 'AGI' part of the name signifies its goal of measuring progress towards artificial general intelligence.
The term ARC-AGI-3 is appearing now because recent AI advancements, like OpenAI's GPT-6 Astra, are claiming record-breaking performance on such challenging reasoning benchmarks. This highlights the ongoing race among leading AI labs to demonstrate progress in abstract reasoning, a key indicator for artificial general intelligence.
- GPT-6 Astra Saturates ARC-AGI-3, Tesla's $30K Cybercab Floods Austin, Anthropic Proves Fermat's Last Theorem | EP #286Anthropic · Moonshots with Peter Diamandis ·
- GPT-6 Astra: A new generation of intelligenceOpenAI · OpenAI Research ·
- OpenAI’s “generational leap” with GPT-6 AstraOpenAI · The Rundown AI ·
- 1 Billion ChatGPT usersOpenAI · Ben's Bites ·
- How enabling two settings tripled our scores on the ARC-AGI-3 benchmarkOpenAI · OpenAI News ·
- Two AI Models Set to “stir government urgency”, But Will This Challenge Undo Them?Anthropic · AI Explained ·
- #235 - Sonnet 4.6, Deep-thinking tokens, Anthropic vs PentagonAnthropic · Last Week in AI ·
- Gemini 3.1 Pro: A smarter model for your most complex tasksGoogle · Google DeepMind Blog ·
- OpenAI Acquires OpenClaw, 400x Cost Collapse, & Why India Wins the Talent War | EP #231OpenAI · Moonshots with Peter Diamandis ·
- OpenAI ChatGPT
- OpenAI's latest models are being evaluated against benchmarks like ARC-AGI-3 to showcase their advanced reasoning capabilities.
- MMLU
- This benchmark tests broad knowledge across many subjects, whereas ARC-AGI
- HellaSwag
- This benchmark tests commonsense reasoning about everyday situations,
- BIG-Bench
- A large, diverse collection of tasks designed to probe AI capabilitie
The ARC benchmark was introduced by the Allen Institute for AI (AI2) to push AI beyond pattern recognition towards more general reasoning abilities. Source ↗
You can try to find online versions or simplified examples of abstract reasoning puzzles. For instance, look for visual pattern completion tests or logic grid puzzles. Then, try solving them yourself and compare your approach to how you imagine an AI might tackle them, focusing on identifying the underlying rules.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
- TemplateEvals CSV prompt list for agent skillsA CSV file template for listing prompts used in evaluating agent skills. · 20 min
- HeuristicDouble coding agent task completion with docs linksProviding coding agents with a documentation link significantly increases their task completion rate. · 2 min
Questions people ask
- What kind of problems are in ARC-AGI-3?
- ARC-AGI-3 contains problems that require identifying abstract patterns and rules from examples, then applying those rules to solve new, unseen problems, often involving visual or logical transformations.
- Why is ARC-AGI-3 important for AI development?
- It's important because it tests a type of intelligence that is closer to human reasoning and problem-solving, which is a key goal for developing more capable and general AI systems.
First seen on Subvolts Feb 18, 2026. Updated Oct 5, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.