{"slug": "arc-agi", "name": "ARC-AGI", "category": "ai", "category_name": "AI tools and apps", "description": "", "recommenders": 3, "commenters": 3, "warners": 0, "score": 2.518, "rising": -0.98, "first": "2025-11-03", "latest": "2026-04-02", "why": "ARC-AGI was specifically designed to be resistant to pre-training", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 3, "neutral": 0, "compare": 0, "warn": 0}, "spark": [1, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "ARC-AGI was specifically designed to be resistant to pre-training", "kind": "", "icon": true, "domain": "arcprize.org", "github_repo": null, "official_url": "https://arcprize.org", "pricing": "", "aliases": ["ARC-AGI"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 761905184, "item_id": 47619074, "url": "https://news.ycombinator.com/item?id=47619074", "by": "nabakin", "date": "2026-04-02", "version": null, "stance": "recommend", "quote": "I know Arc AGI 2 has a private test set and they have a good amount of results", "excerpt": "I know Arc AGI 2 has a private test set and they have a good amount of results[0] but it's not a conventional benchmark. Looking around, SWE Rebench seems to have decent protection against training data leaks[1]. Kagi has one that is fully private[2]. One on HuggingFace that claims to be fully private[3]. SimpleBench[4]. HLE has a private test set apparently[5]. LiveBench[6]. Scale has some private benchmarks but not", "context": "", "use_case": "", "switched_from": "", "story": "Google releases Gemma 4 open models"}, {"id": 751937744, "item_id": 46996109, "url": "https://news.ycombinator.com/item?id=46996109", "by": "mrandish", "date": "2026-02-12", "version": null, "stance": "recommend", "quote": "ARC-AGI was specifically designed to be resistant to pre-training", "excerpt": "> does leak per definition. As a measure focused solely on fluid intelligence, learning novel tasks and test-time adaptability, ARC-AGI was specifically designed to be resistant to pre-training - for example, unlike many mathematical and programming test questions, ARC-AGI problems don't have first order patterns which can be learned to solve a different ARC-AGI problem. The ARC non-profit foundation has private vers", "context": "", "use_case": "AI benchmark", "switched_from": "", "story": "Gemini 3 Deep Think"}, {"id": 732879696, "item_id": 45804981, "url": "https://news.ycombinator.com/item?id=45804981", "by": "xnx", "date": "2025-11-03", "version": null, "stance": "recommend", "quote": "This is pretty much exactly what https://arcprize.org/arc-agi is working on.", "excerpt": "> what is something that only an actually thinking entity can do that LLMs can't? This is pretty much exactly what https://arcprize.org/arc-agi is working on.", "context": "", "use_case": "", "switched_from": "", "story": "The Case That A.I. Is Thinking"}]}], "caveats": [], "who": [], "tags": [], "use_cases": ["ai benchmark"], "compared": [], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=47616361", "title": "Google releases Gemma 4 open models", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=46991240", "title": "Gemini 3 Deep Think", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=45802029", "title": "The Case That A.I. Is Thinking", "mentions": 1}], "sentiment_by_month": {"2025-11": {"recommend": 1}, "2025-12": {}, "2026-01": {}, "2026-02": {"recommend": 1}, "2026-03": {}, "2026-04": {"recommend": 1}, "2026-05": {}, "2026-06": {}, "2026-07": {}, "2026-08": {}, "2026-09": {}, "2026-10": {}}, "versions": [], "version_unnamed": 0, "family_ref": null, "runs_on": null, "lines": [], "quotes": [{"id": 761905184, "item_id": 47619074, "url": "https://news.ycombinator.com/item?id=47619074", "by": "nabakin", "date": "2026-04-02", "version": null, "stance": "recommend", "quote": "I know Arc AGI 2 has a private test set and they have a good amount of results", "excerpt": "I know Arc AGI 2 has a private test set and they have a good amount of results[0] but it's not a conventional benchmark. Looking around, SWE Rebench seems to have decent protection against training data leaks[1]. Kagi has one that is fully private[2]. One on HuggingFace that claims to be fully private[3]. SimpleBench[4]. HLE has a private test set apparently[5]. LiveBench[6]. Scale has some private benchmarks but not", "context": "", "use_case": "", "switched_from": "", "story": "Google releases Gemma 4 open models"}, {"id": 751937744, "item_id": 46996109, "url": "https://news.ycombinator.com/item?id=46996109", "by": "mrandish", "date": "2026-02-12", "version": null, "stance": "recommend", "quote": "ARC-AGI was specifically designed to be resistant to pre-training", "excerpt": "> does leak per definition. As a measure focused solely on fluid intelligence, learning novel tasks and test-time adaptability, ARC-AGI was specifically designed to be resistant to pre-training - for example, unlike many mathematical and programming test questions, ARC-AGI problems don't have first order patterns which can be learned to solve a different ARC-AGI problem. The ARC non-profit foundation has private vers", "context": "", "use_case": "AI benchmark", "switched_from": "", "story": "Gemini 3 Deep Think"}, {"id": 732879696, "item_id": 45804981, "url": "https://news.ycombinator.com/item?id=45804981", "by": "xnx", "date": "2025-11-03", "version": null, "stance": "recommend", "quote": "This is pretty much exactly what https://arcprize.org/arc-agi is working on.", "excerpt": "> what is something that only an actually thinking entity can do that LLMs can't? This is pretty much exactly what https://arcprize.org/arc-agi is working on.", "context": "", "use_case": "", "switched_from": "", "story": "The Case That A.I. Is Thinking"}]}