核心要点
- OpenAI 正在开发一个名为“Astra”的新 AI 模型系列，该系列通过协调多个智能体协同工作，旨在处理长时间运行的任务和复杂问题。
- CEO Sam Altman 已在华盛顿特区展示了 Astra。这些模型目前正在测试中，并将成为首批经过美国计划中的政府审查流程的模型，该流程要求在公开发布前获得官方批准。
- 该项目反映了 OpenAI 更宏大的雄心：构建能够连续数小时甚至数天持续处理问题的 AI 系统。
更新——2026年8月1日
- 新增 Astra 公告及数学论文。
OpenAI 已发布其数学报告，首次正式确认 Astra 这一名称。该公司表示，其“下一代主要模型系列”的内部版本 Astra 解决了数学和理论计算机科学领域的十个开放问题。数学家们至少在十年内未在这些问题上取得任何进展，而在大多数情况下，停滞时间远不止十年。
这些成果涵盖的领域从高维几何、编码理论，到群论、量子复杂性、格密码学以及极值组合学。其中一项证明确立了非 sofic 群的存在性，解决了群论中的一个重大开放问题。
曼彻斯特大学数学家、erdosproblems.com 网站运营者 Thomas Bloom 在 X 上称这些成果为“重大新闻”。他认为这些成果比 5 月发表的单位距离猜想反例更具意义。“也许比不上单位距离猜想证明本身的分量，但就构造而言，这确实重大，”Bloom 写道。
Bloom 还驳斥了 AI 正在取代数学家的观点，认为当 AI 借鉴了一个多世纪的数学理论、由数学家构建、并基于数学家们写下的所有内容进行训练时，这种说法几乎站不住脚。
参与开发 Astra 所用测试时推理技术的研究人员之一 Noam Brown 在 X 上表示，OpenAI 也曾尝试攻克其他重大难题但未成功。“遗憾的是，（千禧年大奖难题）一个都还没解决，”他写道。克莱数学研究所为七大千禧年大奖难题各悬赏 100 万美元，但自 2000 年奖项公布以来，仅有一道被攻克。Brown 补充道：“不过，我们在每个问题上投入的成本并不高。测试时算力还有很大的提升空间。”他称 Astra 是“科学推理领域的重大一步”。
Astra 的解答按 API 价格计算，成本约为 2000 美元
OpenAI 表示，按 Sol 的 API 价格计算，生成全部十个解答所用的 token 成本约为 2000 美元。模型生成论证后，人类与同一模型协作，将其整理成研究论文。模型还在 Lean 中形式化了每个证明，生成了可机器校验的数学正确性证书，OpenAI 并发布了每个解答背后模型推理过程的逐步解析。
OpenAI 表示，其研究人员协助撰写了论文并形式化了证明，公司对这些内容的准确性负责。然而，数学论证本身出自 Astra。
该公司认为，将完全由 AI 生成的证明归于人类作者身份，既会歪曲该系统的贡献，也会歪曲真正人类智力工作的本质，并援引《莱顿 AI 与数学宣言》作为 AI 辅助研究中署名归属的参考依据。
据报道，OpenAI 正在构建 Astra——一个旨在连续工作数小时甚至数天解决难题的模型系列
OpenAI 正在开发一个暂定名为“Astra”的新模型系列，其设计目标是在长时间运行任务上的能力远超该公司迄今已发布的任何产品。
本周，CEO Sam Altman 在华盛顿特区向政界人士和监管机构演示了 Astra。OpenAI 强调该系统能够在较长时间内协调多个智能体，以攻克特别棘手的问题。该公司指出，复杂项目和高等数学是潜在的应用场景。The Information 援引三位知情人士的消息报道了这一细节。
据报道，Astra 将作为全新模型类别，与 OpenAI 现有的 Sol、Terra 和 Luna 模型家族并列。它究竟会以 GPT-6 的形式发布，还是作为 GPT-5 系列中的一个变体（类似 GPT 5.7）推出，目前尚未决定。发布日期也尚未确定。
OpenAI 还计划很快发布一份报告，展示该公司如何利用其最先进的 AI 解决了十个此前悬而未决的数学问题。其目的是展示当前模型已经具备的能力。
Astra 将成为美国新监管框架下首个接受测试的模型
据 The Information 报道，这些模型已进入测试阶段。它们预计将成为首批接受特朗普政府计划推出的新 AI 框架审查的模型，该框架要求 AI 模型在公开发布前提交给联邦政府。政府的目标是在本周结束前敲定该框架。
一个关键问题是，在长时间运行的工作流程中，随着上下文不断增长，模型能否避免错误累积，并在流程偏离轨道时自我纠正。这仍然是当今智能体系统的一大弱点。像 Astra 这样的多智能体架构在处理紧密关联的任务（如规划）时，表现也可能更差，因为协调开销和错误累积会抵消掉任何收益。
长期目标是实现自主 AI 研究。
Astra 的传闻与该公司早前的表态相吻合。首席科学家 Jakub Pachocki 去年夏天在 OpenAI 官方播客上表示，公司希望构建能够连续数小时或数天处理一个问题的 AI 系统。当前系统往往局限于短时任务，但 OpenAI 希望模型能够在更长的时间跨度内进行规划、推理和实验。去年年底，该公司甚至提出了这样一个问题：如何思考那些能够解决人类需要数百年才能完成的任务的系统。
到 2028 年 3 月，OpenAI 希望拥有一名完全自主的 AI 研究员，能够独立运行研究项目，而该系统也将依赖于长时间运行的 AI 进程。最早在今年 9 月，该公司计划推出一款具备研究实习生级技能的 AI 系统，这将大幅加快人类科学家的研究速度。Astra 最终可能就是这个系统。
Pachocki 还表示，这些系统将需要远多于当前的算力。OpenAI 的长期基础设施规划也反映了这一雄心。这家初创公司的收入能否以足够快的速度增长来支撑如此大规模的建设，仍是一个悬而未决的问题。
不炒作的 AI 新闻——由人工精选
The Information
Key Points
- OpenAI is working on a new AI model family called "Astra," built to handle long-running tasks and complex problems by coordinating multiple agents working together.
- CEO Sam Altman has already showcased Astra in Washington, D.C. The models are currently being tested and will be the first to go through a planned U.S. government review process that requires official approval before public release.
- The project reflects OpenAI's broader ambition to build AI systems capable of working on problems continuously for hours or even days at a time.
Update – Aug 1, 2026
- Added Astra announcement and math paper.
OpenAI has released its math report, officially confirming the Astra name for the first time. The company says an internal version of Astra, its "next major model family," solved ten open problems in math and theoretical computer science. Mathematicians had made no progress on any of them for at least a decade, and much longer in most cases.
The results cover fields ranging from high-dimensional geometry and coding theory to group theory, quantum complexity, lattice cryptography, and extremal combinatorics. One proof establishes the existence of non-sofic groups, resolving a major open question in group theory.
Thomas Bloom, a University of Manchester mathematician who runs erdosproblems.com, called the results "big news" on X. He considers them more significant than the counterexample to the unit distance conjecture published in May. "Maybe not bigger than a proof of unit distance would have been, but in terms of constructions, this is big," Bloom wrote.
Bloom also rejected the idea that AI is replacing mathematicians, arguing that the claim makes little sense when the AI draws on more than a century of mathematical theory, was built by mathematicians, and was trained on everything mathematicians have ever written.
Noam Brown, one of the researchers behind the test-time reasoning technology used by Astra, said on X that OpenAI had also tried and failed to crack other major problems. "Sadly, no Millennium Prize Problems (yet)," he wrote. The Clay Mathematics Institute offers $1 million for solving each of the seven Millennium Prize Problems, but only one has been solved since the prizes were announced in 2000. Brown added, "But also, we didn't spend a lot on each problem. It's possible to push test-time compute much further." He called Astra a "major step for scientific reasoning."
Astra's solutions would cost about $2,000 at API rates
OpenAI says the tokens used to generate all ten solutions would have cost about $2,000 at Sol's API rates. After the model produced its arguments, humans worked with the same model to turn them into research papers. The model also formalized each proof in Lean, creating machine-checkable certificates of mathematical correctness, and OpenAI published a walkthrough of the model's reasoning process for each solution.
OpenAI said its researchers helped prepare the papers and formalize the proofs, and that the company takes responsibility for their accuracy. The mathematical arguments themselves, however, came from Astra.
The company argued that claiming human authorship for a proof generated entirely by AI would misrepresent both the system's contribution and the nature of genuine human intellectual work, pointing to the Leiden Declaration on AI and Mathematics as a reference for how credit should be assigned in AI-assisted research.
OpenAI is reportedly building Astra, a model family designed to work on problems for hours or days
OpenAI is working on a new model family tentatively called "Astra" that's meant to be far more capable at long-running tasks than anything the company has shipped so far.
CEO Sam Altman demoed Astra to politicians and regulators in Washington, D.C., this week. OpenAI stressed the system's ability to coordinate multiple agents over extended periods to tackle especially hard problems. The company pointed to complex projects and advanced math as potential use cases. The Information reported the details, citing three people familiar with the plans.
According to the report, Astra would form a new model class alongside OpenAI's existing Sol, Terra, and Luna families. Whether it ships as GPT-6 or as a variant within the GPT-5 line, something like GPT 5.7, hasn't been decided yet. There's no release date either.
OpenAI also plans to publish a report soon showing how the company used its most advanced AI to solve ten previously unsolved math problems. The goal is to show what its current models can already do.
Astra would be the first model tested under a new US regulatory framework
The models are already in testing, according to The Information. They're expected to be the first to go through the Trump administration's planned new AI framework, which would require AI models to be submitted to the federal government before public release. The administration aims to finalize the framework by the end of this week.
One key question is whether the models can avoid compounding errors during long-running workflows and correct themselves when a process drifts off course as the context keeps growing. That remains a major weakness in today's agentic systems. Multi-agent setups like Astra can also perform worse on tightly linked tasks such as planning because coordination overhead and compounding errors can wipe out any gains.
The long-term goal is autonomous AI research
The Astra rumors line up with earlier statements from the company. Chief Scientist Jakub Pachocki said on OpenAI's official podcast last summer that the company wants to build AI systems that can work on a problem for hours or days. Current systems are often limited to short tasks, but OpenAI wants models that can plan, reason, and experiment over longer time horizons. Late last year, the company even raised the question of how to think about systems that could solve tasks a human would need centuries to complete.
By March 2028, OpenAI wants to have a fully autonomous AI researcher that can run research projects on its own, and that system would also depend on long-running AI processes. As early as this September, the company plans to have an AI system with research-intern-level skills that would significantly speed up human scientists. Astra could end up being that system.
Pachocki also said these systems will need far more compute. OpenAI's long-term infrastructure plans reflect that ambition. Whether the startup's revenue grows fast enough to fund that massive buildout remains an open question.
AI News Without the Hype – Curated by Humans
The Information