O conteúdo está em inglês; a navegação e os filtros mantêm o idioma selecionado.

OpenAI previews next-generation model codenamed Astra with ten mathematical advances

Posts linked to OpenAI and several media reports describe Astra as an internal next-generation model that produced ten advances in mathematics and theoretical computer science. The reported work spans group theory, geometry, cryptography, complexity, and combinatorics, but independent verification is still pending.

Publicado em 06/08/2026, 09:21:33Última atualização 06/08/2026, 09:21:33Publicação editorial

Available reports portray Astra as a possible model family designed for long-running, multi-agent collaboration on difficult tasks. Material associated with OpenAI researcher Sébastien Bubeck refers to ten new results, including work involving non-sofic groups, von Neumann algebras, high-dimensional sphere packing, coding theory, and quantum complexity.

Secondary reports put the total inference cost at roughly $2,000 and say the results will be accompanied by Lean certificates and explanations of the reasoning process. They also claim that CEO Sam Altman demonstrated Astra to policymakers in Washington. OpenAI has not publicly established a launch date, commercial pricing, or whether the system will be branded as GPT-6 or as part of the GPT-5 line.

The wording matters: the primary-looking material describes “advances” and “results,” while some coverage calls them definitive solutions to ten long-standing problems or even Fields Medal-level breakthroughs. Because the evidence is largely based on reposts and secondary reporting rather than a fully accessible official paper and independent peer review, the claims should be treated as provisional.

Evidências das fontes 6

A verificação de fontes não garante exatidão permanente. Considere a fonte original, a data e as correções posteriores.

OpenAI新模型Astra曝！攻克10道數學難題、多AI代理協作挑戰AGIhk.finance.yahoo.com · Complementar

... 鉅亨網 # OpenAI新模型Astra曝！攻克10道數學難題、多AI代理協作挑戰AGI OpenAI新模型Astra曝！攻克10道數學難題、多AI代理協作挑戰AGI · 鉅亨網鉅亨網編譯莊閔棻 OpenAI 正在開發代號「Astra」的全新模型家族，主打多個人工智慧（AI）代理長時間協同運作、共同解決高難度問題的能力。這項技術若順利問世，將是該公司邁向通用人工智慧（AGI）的重要里程碑，同時也可能成為美國川普政府新版 AI 監管框架上路後，第一批接受審查的產品。根據《The Information》引述三名知情人士報導，OpenAI 執行長奧特曼（Sam Altman）本週親自前往華盛頓特區，向政策制定者與監管官員展示 Astra 的實際運作。簡報中特別強調該模型能調度多個智能體長時間分工合作，處理從專案管理到高階數學證明等複雜任務。與此同時，OpenAI 也發布一份報告，公布其最先進 AI 系統成功攻克十道長年懸而未解的數學難題，範圍橫跨高維幾何、程式設計理論、群論、量子複雜性與極值組合學等尖端領域。這些問題各自懸置學界超過十年，而 OpenAI 僅耗費約兩千美元的運算成本便完成全部求解。今年一月加入 OpenAI 的研究員 Lijie Chen 在社群平台上證實，這十道題目的證明正是出自旗艦模型 Astra 之手。根據知情人士說法，Astra 將自成一個獨立的模型類別，與 OpenAI 現有的 Sol、Terra、Luna 系列並列。目前公司內部尚未拍板該模型上市時間，也還沒決定其命名方式，究竟會掛牌為外界期待已久的 GPT-6，或僅作為 GPT-5 系列的子版本（如 GPT-5.7）發布，仍是未知數。