{"rewrite":{"id":"r_045ae61c7fedc4c11f99eab0","clusterId":"c_50bee5ab3221faef9b534e79","slug":"openai-report-shows-median-researcher-ai-spend-tops-600-a-day","model":"deepseek-v4-flash","headline":"OpenAI Report Shows Median Researcher AI Spend Tops $600 a Day","summary":"OpenAI published a report on September 6, 2026 detailing how its own researchers use AI. Median daily token spending passed $600 (about 94,000 yen) as of August 2026, with the top 10 percent above $7,000. Researcher output tokens rose 124 times between November 2025 and August 2026. AI agent working hours passed human hours in June 2026 and reached 3.14 times human hours by mid-August.","whyItMatters":"The report puts OpenAI's own operations behind its transparency stance on self-improving AI, showing AI agents now do the majority of research labor inside the company.","webCardHtml":"\u003cp\u003eAbout 70 percent of OpenAI researchers were running four or more AI agents at once as of August 2026, according to the report, a share that has climbed since the data series began in late 2025.\u003c/p\u003e\u003cp\u003eThe report also tracks task difficulty. For jobs that take a human 64 to 128 hours, 16 percent were completed without any human intervention, 51 percent needed at least one intervention, 26 percent failed outright, and 7 percent hit tool errors.\u003c/p\u003e\u003cp\u003eOpenAI frames the release as part of its transparency stance on self-improving AI: it argues developers should be required to publish progress on such systems, and it is publishing its own numbers without being required to. The report notes GPT-6 Astra was already in internal research use before its public announcement, and that Astra-related reinforcement learning declined after the attack on Hugging Face.\u003c/p\u003e","blueskyPost":"OpenAI published its internal numbers on AI use by its own researchers: median daily token spend over $600, top decile over $7,000, and AI agent hours now 3.14x human hours. The company says it is publishing voluntarily as part of its transparency stance on self-improving AI.","twitterPost":"OpenAI's own report: median researcher AI spend tops $600/day, top decile over $7,000. Output tokens up 124x since Nov 2025. AI agents now work 3.14x human hours. OpenAI publishes voluntarily as part of its self-improving AI transparency stance.","threadsPost":null,"newsletterBlurb":"OpenAI released a report showing how its own researchers use AI: median daily token spending above $600, output tokens up 124 times since late 2025, and AI agents now working more than three times as many hours as humans. The company frames the release as voluntary transparency on its path toward self-improving AI.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260907-ai-use-inside-openai/\",\"title\":\"OpenAI研究員によるAI利用の中央値は「1日9万円超え」で上位10％は「1日100万円超え」\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":5361,"outputTokens":3174,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1788751801,"createdAt":"2026-09-07T03:27:15.000Z","publishedAt":"2026-09-07T03:28:25.000Z","updatedAt":"2026-09-07T03:28:25.000Z"},"cluster":{"id":"c_50bee5ab3221faef9b534e79","canonicalTitle":"OpenAI研究員によるAI利用の中央値は「1日9万円超え」で上位10％は「1日100万円超え」","representativeArticleId":"a_2307ba0e0dbd2c1ae3e47476","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[],\"studios\":[],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-09-07T02:42:00.000Z","lastSeenAt":"2026-09-07T02:42:00.000Z","updatedAt":"2026-09-07T03:28:20.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260907-ai-use-inside-openai/","title":"OpenAI研究員によるAI利用の中央値は「1日9万円超え」で上位10％は「1日100万円超え」"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":[],"studios":[],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":null}