{"source":"AIMap","schema":"benchmark/v1","updated":"2026-10-22","benchmark":{"id":"arc-agi-2","name":"ARC-AGI-2","category":"reasoning","unit":"%","higherIsBetter":true,"status":"active","sources":["arcprize"],"primaryVariant":"semi-private","url":"https://arcprize.org/leaderboard","year":2025,"desc":{"en":"Harder second version of ARC-AGI (2025), designed so that brute-force search is not enough.","zh-hans":"ARC-AGI 第二版（2025），更难，单靠暴力搜索无法解出。"},"measures":{"en":"% solved on the semi-private set, with cost per task.","zh-hans":"半私有题集解题率及每题成本。"},"limits":{"en":"Only the top entries are cited (terms). Systems with large test-time compute can buy accuracy with money.","zh-hans":"按条款只引用前几名。大算力推理系统可以用钱换分数。"},"contamination":"low","contaminationNote":{"en":"Semi-private set.","zh-hans":"半私有题集。"},"stats":{"results":10,"models":4,"latest":"2026-10-22","bySourceType":{"official":10}}},"sources":[{"id":"arcprize","name":"ARC Prize leaderboard","license":"ARC Prize Terms: no copying/aggregation for commercial purposes without permission","policy":"limited","attribution":"ARC Prize Foundation, arcprize.org/leaderboard","site":"https://arcprize.org/leaderboard"}],"results":[{"model":"gpt-6-astra","modelName":"GPT-6 Astra","provider":"OpenAI","rawName":"GPT-6 Astra (Max)","variant":"semi-private","setting":"Semi-private set; $1.12/task; CoT","score":95,"unit":"%","rank":1,"rankComputedBySite":false,"date":"2026-10-22","sourceType":"official","source":"arcprize","sourceUrl":"https://arcprize.org/leaderboard","collected":"2026-10-10"},{"model":"gpt-6-1-sol","modelName":"GPT-6.1 Sol","provider":"OpenAI","rawName":"GPT-6.1 Sol (Max)","variant":"semi-private","setting":"Semi-private set; $0.25/task; CoT","score":94.2,"unit":"%","rank":2,"rankComputedBySite":false,"date":"2026-10-22","sourceType":"official","source":"arcprize","sourceUrl":"https://arcprize.org/leaderboard","collected":"2026-10-10"},{"model":"gpt-6-astra","modelName":"GPT-6 Astra","provider":"OpenAI","rawName":"GPT-6 Astra (XHigh)","variant":"semi-private","setting":"Semi-private set; $0.83/task; CoT","score":93.3,"unit":"%","rank":3,"rankComputedBySite":false,"date":"2026-10-22","sourceType":"official","source":"arcprize","sourceUrl":"https://arcprize.org/leaderboard","collected":"2026-10-10"},{"model":"claude-opus-5-5","modelName":"Claude Opus 5.5","provider":"Anthropic","rawName":"Claude Opus 5.5 (High)","variant":"semi-private","setting":"Semi-private set; $0.41/task; CoT","score":93.3,"unit":"%","rank":4,"rankComputedBySite":false,"date":"2026-10-22","sourceType":"official","source":"arcprize","sourceUrl":"https://arcprize.org/leaderboard","collected":"2026-10-10"},{"model":"gpt-5-6-sol","modelName":"GPT-5.6 Sol","provider":"OpenAI","rawName":"GPT-5.6 Sol (Max)","variant":"semi-private","setting":"Semi-private set; $1.44/task; CoT","score":92.5,"unit":"%","rank":5,"rankComputedBySite":false,"date":"2026-10-22","sourceType":"official","source":"arcprize","sourceUrl":"https://arcprize.org/leaderboard","collected":"2026-10-10"},{"model":"claude-opus-5-5","modelName":"Claude Opus 5.5","provider":"Anthropic","rawName":"Claude Opus 5.5 (XHigh)","variant":"semi-private","setting":"Semi-private set; $0.67/task; CoT","score":92.5,"unit":"%","rank":6,"rankComputedBySite":false,"date":"2026-10-22","sourceType":"official","source":"arcprize","sourceUrl":"https://arcprize.org/leaderboard","collected":"2026-10-10"},{"model":"gpt-6-astra","modelName":"GPT-6 Astra","provider":"OpenAI","rawName":"GPT-6 Astra (High)","variant":"semi-private","setting":"Semi-private set; $0.67/task; CoT","score":92.1,"unit":"%","rank":7,"rankComputedBySite":false,"date":"2026-10-22","sourceType":"official","source":"arcprize","sourceUrl":"https://arcprize.org/leaderboard","collected":"2026-10-10"},{"model":"gpt-6-astra","modelName":"GPT-6 Astra","provider":"OpenAI","rawName":"GPT-6 Astra (Medium)","variant":"semi-private","setting":"Semi-private set; $0.48/task; CoT","score":92.1,"unit":"%","rank":8,"rankComputedBySite":false,"date":"2026-10-22","sourceType":"official","source":"arcprize","sourceUrl":"https://arcprize.org/leaderboard","collected":"2026-10-10"},{"model":"claude-opus-5-5","modelName":"Claude Opus 5.5","provider":"Anthropic","rawName":"Claude Opus 5.5 (Max)","variant":"semi-private","setting":"Semi-private set; $1.85/task; CoT","score":91.7,"unit":"%","rank":9,"rankComputedBySite":false,"date":"2026-10-22","sourceType":"official","source":"arcprize","sourceUrl":"https://arcprize.org/leaderboard","collected":"2026-10-10"},{"model":"gpt-6-1-sol","modelName":"GPT-6.1 Sol","provider":"OpenAI","rawName":"GPT-6.1 Sol (XHigh)","variant":"semi-private","setting":"Semi-private set; $0.18/task; CoT","score":91.7,"unit":"%","rank":10,"rankComputedBySite":false,"date":"2026-10-22","sourceType":"official","source":"arcprize","sourceUrl":"https://arcprize.org/leaderboard","collected":"2026-10-10"}]}