{"rewrite":{"id":"r_11e0160e4624416bfdad59b2","clusterId":"c_76bf84f11717084f516d8fe4","slug":"gpt-6-astra-scores-99-9-on-arc-agi-3-as-openai-stops-short-of-agi-claim","model":"deepseek-v4-flash","headline":"GPT-6 Astra Scores 99.9% on ARC-AGI-3 as OpenAI Stops Short of AGI Claim","summary":"OpenAI announced GPT-6 Astra on September 3, reporting 99.9 percent on ARC-AGI-3, a benchmark that tests whether a model can explore an unexplained environment, infer rules, and plan. Astra cleared 96 percent of stages in fewer operations than the human median, with average operations 51.7 percent below human. OpenAI says the result is not AGI: the score used a custom Provider Adapter, and the Standard harness topped out at 62.7 percent.","whyItMatters":"Astra nearly saturated ARC-AGI-3, but the top score relied on a custom harness and ARC Prize calls benchmark saturation no proof of AGI, so the milestone is real while the AGI claim stays qualified.","webCardHtml":"\u003cp\u003eThe rest of the benchmark sheet is mixed. Astra scored 57.9 percent on Terminal-Bench 4.0, ahead of Claude Fable 5.1\u0026#39;s 55.8 percent and GPT-5.6 Sol\u0026#39;s 37.3 percent, and 64.6 percent on Terminal-Bench Science 0.1. It reached 97.6 percent on FrontierMath Tier 4.\u003c/p\u003e\u003cp\u003eThe cyber benchmark is the one OpenAI flags. Astra hit 100 percent on ExploitBench, and OpenAI rates it at Critical cyber capability under its safety standards, with restrictions planned for requests that could lead to advanced attacks.\u003c/p\u003e\u003cp\u003eARC Prize pushed back on reading the result as a finish line. The benchmark uses a closed environment and is a limited evaluation, and saturation is not proof of AGI achievement. OpenAI agrees: the 99.9 percent figure came through a Provider Adapter that holds inference state for OpenAI, while the Standard harness reached 62.7 percent at best.\u003c/p\u003e","blueskyPost":"OpenAI's GPT-6 Astra scored 99.9% on ARC-AGI-3, clearing 96% of stages in fewer operations than the human median. OpenAI and ARC Prize both say that is not AGI: the score used a custom harness, and the Standard harness topped out at 62.7%.","twitterPost":"GPT-6 Astra hit 99.9% on ARC-AGI-3, solving 96% of stages with 51.7% fewer average operations than humans. OpenAI and ARC Prize both caution it is not AGI: the top score used a custom Provider Adapter, and the Standard harness reached 62.7%.","threadsPost":null,"newsletterBlurb":"OpenAI announced GPT-6 Astra on September 3, reporting 99.9 percent on ARC-AGI-3, a benchmark for exploring unexplained environments and inferring rules. Astra cleared 96 percent of stages in fewer operations than the human median, but OpenAI and ARC Prize both caution the result is not proof of AGI, since the top score used a custom harness and the Standard harness reached 62.7 percent. The model also hit 100 percent on the ExploitBench cyber benchmark, earning a Critical cyber rating under OpenAI's safety standards.","attributionJson":"[{\"source\":\"ASCII.jp\",\"url\":\"https://ascii.jp/elem/000/004/432/4432209/?rss\",\"title\":\"OpenAI GPT-6 Astra's Remarkable Evolution: Power to Challenge the Unknown Surpasses Humans\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":6178,"outputTokens":4728,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":null,"createdAt":"2026-09-04T01:15:36.000Z","publishedAt":"2026-09-04T01:16:40.000Z","updatedAt":"2026-09-04T01:16:40.000Z"},"cluster":{"id":"c_76bf84f11717084f516d8fe4","canonicalTitle":"OpenAI「GPT-6 Astra」驚異の進化　“未知に挑む力”人間超える","representativeArticleId":"a_9c8977461095068a9e5fdd08","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[],\"studios\":[],\"people\":[],\"type\":\"announcement\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-09-04T00:35:00.000Z","lastSeenAt":"2026-09-04T00:35:00.000Z","updatedAt":"2026-09-04T01:16:40.000Z"},"attribution":[{"source":"ASCII.jp","url":"https://ascii.jp/elem/000/004/432/4432209/?rss","title":"OpenAI「GPT-6 Astra」驚異の進化　“未知に挑む力”人間超える"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":[],"studios":[],"people":[],"type":"announcement","domain":"other","is_roundup":false},"keyFacts":null}