# ARC-AGI 22 Hacker News commenters recommended ARC-AGI and 24 criticized it, November 2025 to November 2026. It was recommended in 8 of 13 months. **Verdict:** Puzzles designed to be easy for humans and resistant to training-to-the-test, used to compare AI reasoning. Critics say it predicts little and labs game it. ## Why people recommend it - Good, clever, better designed benchmark (4 comments): "This is a good and clever benchmark and a worthy successor to the previous two." ([source](https://news.ycombinator.com/item?id=47523704)) "ARC-AGI has thought a fair deal about this and has a good definition here" ([source](https://news.ycombinator.com/item?id=49115047)) "I tend to prefer the ARC-AGI benchmarks for the most part." ([source](https://news.ycombinator.com/item?id=47735071)) - Reasoning-heavy puzzles people enjoy (4 comments): "I tried out their human simulation, and it feels very reasoning heavy." ([source](https://news.ycombinator.com/item?id=47911148)) "it feels very reasoning heavy" ([source](https://news.ycombinator.com/item?id=47911656)) "pretty fun to run as a human" ([source](https://news.ycombinator.com/item?id=49048095)) - Resistant to training-to-the-test (2 comments): "ARC-AGI was designed specifically for evaluating deeper reasoning in LLMs, including being resistant to LLMs 'training to the test'." ([source](https://news.ycombinator.com/item?id=46236995)) "ARC-AGI was specifically designed to be resistant to pre-training" ([source](https://news.ycombinator.com/item?id=46996109)) - Tests tasks humans find easy (2 comments): "the tests are designed to stress frontier models in tasks that humans are uniquely good at" ([source](https://news.ycombinator.com/item?id=49576858)) "ARC-AGI-3 was more of a failure in that regard than -1 or -2" ([source](https://news.ycombinator.com/item?id=49558272)) ## What people warn about - Doesn't measure real intelligence (4 comments): "I think it is a nothingburger of a benchmark: lots of grand claims about AGI, but very little predictive power or practical utility." ([source](https://news.ycombinator.com/item?id=47542783)) "ARC-AGI is much more bounded than they seem to think. It correlates with intelligence, but doesn't imply it." ([source](https://news.ycombinator.com/item?id=46237068)) "It has nothing to do wirh some AGI testing. It is just some kind of useless weird game." ([source](https://news.ycombinator.com/item?id=47541901)) - Name overstates what it measures (2 comments): "I'm glad we have the ARC series of tests. But they have "AGI" right in the name of the test." ([source](https://news.ycombinator.com/item?id=47005871)) "It's completely misnamed. It should be called useless visual puzzle benchmark 2." ([source](https://news.ycombinator.com/item?id=46992958)) - Benchmark gaming and lab marketing (3 comments): "OpenAI and Gemini in particular use this as marketing material. You can correlate the benchmark release with stock price increase." ([source](https://news.ycombinator.com/item?id=48145713)) "benchmarks like this are often flawed because leading model labs frequently participate in 'benchmarkmaxxing'" ([source](https://news.ycombinator.com/item?id=46992497)) "every benchmark will be compromised with providers being able tweak things sufficiently to game" ([source](https://news.ycombinator.com/item?id=49556414)) - Simple puzzles, unsuited to LLMs (3 comments): "It seems like more of a simple optimization problem (think Sokoban) than anything approaching fluid intelligence." ([source](https://news.ycombinator.com/item?id=49039413)) "ARC-AGI is a terrible benchmark for testing LLMs because LLMs are not made, trained, or tuned for playing games." ([source](https://news.ycombinator.com/item?id=49048788)) "Anyone taking a single look at the ARC-AGI "challenges" can see things a 5 year old could reasonably solve." ([source](https://news.ycombinator.com/item?id=49561380)) ## Alternatives hackers recommend - [Artificial Analysis](https://hackerapproved.com/p/artificial-analysis.md): 62 recommend - [LMArena](https://hackerapproved.com/p/lmarena.md): 39 recommend - [OCR Arena](https://hackerapproved.com/p/ocr-arena.md): 11 recommend - [Bullshit Benchmark](https://hackerapproved.com/p/bullshit-benchmark.md): 9 recommend - [GenAI Showdown](https://hackerapproved.com/p/genai-showdown.md): 9 recommend ## Cite as Hacker Approved, "ARC-AGI", https://hackerapproved.com/p/arc-agi-3/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).