The ARC Prize organization designs benchmarks which are specifically crafted to demonstrate tasks that humans complete easily, but are difficult for AIs like LLMs, "Reasoning" models, and Agentic frameworks.
ARC-AGI-3 is the first fully interactive benchmark in the ARC-AGI series. ARC-AGI-3 represents hundreds of original turn-based environments, each handcrafted by a team of human game designers. There are no instructions, no rules, and no stated goals. To succeed, an AI agent must explore each environment on its own, figure out how it works, discover what winning looks like, and carry what it learns forward across increasingly difficult levels.
Previous ARC-AGI benchmarks predicted and tracked major AI breakthroughs, from reasoning models to coding agents. ARC-AGI-3 points to what's next: the gap between AI that can follow instructions and AI that can genuinely explore, learn, and adapt in unfamiliar situations.
You can try the tasks yourself here: https://arcprize.org/arc-agi/3
Here is the current leaderboard for ARC-AGI 3, using state of the art models
- OpenAI GPT-5.4 High - 0.3% success rate at $5.2K
- Google Gemini 3.1 Pro - 0.2% success rate at $2.2K
- Anthropic Opus 4.6 Max - 0.2% success rate at $8.9K
- xAI Grok 4.20 Reasoning - 0.0% success rate $3.8K.
(Logarithmic cost on the horizontal axis. Note that the vertical scale goes from 0% to 3% in this graph. If human scores were included, they would be at 100%, at the cost of approximately $250.)
https://arcprize.org/leaderboard
Technical report: https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf
In order for an environment to be included in ARC-AGI-3, it needs to pass the minimum “easy for humans” threshold. Each environment was attempted by 10 people. Only environments that could be fully solved by at least two human participants (independently) were considered for inclusion in the public, semi-private and fully-private sets. Many environments were solved by six or more people. As a reminder, an environment is considered solved only if the test taker was able to complete all levels, upon seeing the environment for the very first time. As such, all ARC-AGI-3 environments are verified to be 100% solvable by humans with no prior task-specific training
I tend to be anti-AI because it doesn't seem to me to be anything other than a super fast regurgitator of data. If a database can be searched for an answer, AI can do that faster than a human. However it doesn't to seem to be able to take some portion of that database, understand it, and then use that information to solve a novel problem.
Well... It cannot even search databases without errors.
LLMs just produce plausible replies in natural languages very quickly and this is useful in certain situations. Sometimes it helps humans getting started with a task, but as it is now, it cannot replace them. As much as the capital class want it, and sink our money into it.
The better setup generate "semantic embeddings" that try to map how data stored relate to each other (by mapping how to it related within in its own weights and biases). That and knowledge graph look ups in which the links between different articles of data are evaluated in the same way.
The very expensive LLM portion really do just give rough aproximations of information language in that setup
Yes, the key thing is it might have extracted useful info from otherwise confusing data, it might have mixed up info from the data incorrectly or it might have just made it up.
So it can be useful, if you can then validate the info provided in more traditional means, but it's dubious as a first pass, and sometimes surprisingly bad when it's a scenario you thought it would work well at.