The new results cover open problems in geometry, cryptography, and complexity.
Pavel Danilyuk
OpenAI announced new results addressing long-standing open problems in mathematics and theoretical computer science. The findings were presented as a set of ten advances in these fields.
OpenAI presented ten advances in mathematics and theoretical computer science.
OpenAI researcher Noam Brown stated that an internal version of the upcoming 'Astra' model family solved the 10 major open problems mentioned, noting it will be a major step for scientific reasoning. Additionally, the company is currently offering GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna models to users, though the official release date for the Astra family remains unknown.
OpenAI researcher Noam Brown revealed that an internal version of the upcoming 'Astra' model family successfully solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science. While the release date for the Astra family remains unknown, Brown noted that the achievement represents a major step for scientific reasoning. Additionally, the company is reportedly offering GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna AI models to users.
OpenAI researcher Noam Brown revealed that an internal version of the upcoming 'Astra' model family solved the 10 major open problems previously mentioned, adding that the release date for the new models remains unknown. Additionally, the company is currently offering GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna models to users.
OpenAI researcher Noam Brown revealed that an internal version of the upcoming 'Astra' model family successfully solved the 10 major open problems previously announced, calling it a major step for scientific reasoning. While the company is currently offering GPT-5.6 Sol, Terra, and Luna models to users, the official release date for the Astra family remains unknown.
OpenAI researcher Noam Brown revealed that an internal version of the upcoming 'Astra' model family solved the ten major open problems previously announced. While the release date for the Astra family remains unknown, Brown noted that the models currently being offered to users include GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna.
OpenAI researcher Noam Brown revealed that an internal version of the upcoming 'Astra' model family solved the ten major open problems, providing Lean certificates for verification. The models achieved breakthroughs in areas such as non-sofic groups, Connes's rigidity conjecture, and problems posed by Paul Erdős, with solution costs estimated at roughly $2,000 via Sol API rates. While Astra is being developed for long-running workloads and agent collaboration, its official release name and date remain unconfirmed.
OpenAI researcher Noam Brown revealed that the upcoming 'Astra' model family, which is being developed for long-running workloads, solved 10 major open problems in fields such as quantum complexity and lattice cryptography by formalizing arguments as Lean certificates. These achievements include the disproof of Connes’s rigidity conjecture and new bounds for high-dimensional sphere packing, resolving problems that had seen no progress for at least a decade. While specific release names like GPT-5.7 or GPT-6 remain undecided, the solutions provided using Astra were estimated to cost approximately $2,000 at Sol API rates.
OpenAI researcher Noam Brown revealed that an internal version of the upcoming 'Astra' model family solved the 10 major open problems previously announced, formalizing every argument as a Lean certificate for mathematical verification. This new model family, which is reportedly built for long-running workloads and agent collaboration, achieved these results in areas such as high-dimensional geometry and lattice cryptography without requiring extraordinary computing power. Additionally, OpenAI has begun publicly releasing the research texts, Lean verification files, and solution process explanations for these solved problems.
Researcher Noam Brown revealed that the upcoming 'Astra' model family solved these 10 problems by formalizing arguments as Lean certificates for mathematical verification. OpenAI has also publicly released the research texts, Lean verification files, and explanations regarding the model's solution processes. Additionally, reports suggest OpenAI is testing a 'ChatGPT for Science' subscription as part of its scientific reasoning advancements.
OpenAI researcher Noam Brown revealed that an internal version of the upcoming 'Astra' model family solved these 10 major open problems, which had seen no progress for at least a decade. The Astra model, designed for long-running workloads and collaborative AI agents, formalized its arguments as Lean certificates and provided examples such as disproving Connes’s rigidity conjecture. Additionally, while the release name remains undecided, OpenAI has already expanded access to the GPT-5.6 model series and recently saw price drops for the Luna and Terra models by 80 percent and 20 percent, respectively.
OpenAI researcher Noam Brown revealed that the upcoming 'Astra' model family achieved these breakthroughs, noting that an internal version solved 10 major open problems in fields such as quantum complexity and lattice cryptography. The model's achievements, which include disproving Connes’s rigidity conjecture and resolving problems posed by Paul Erdős, were formalized as Lean certificates for mathematical verification. Additionally, while the release name and date for Astra remain unknown, The Information confirmed the model family is being built specifically for long-running workloads.
OpenAI researcher Noam Brown revealed that an internal version of the upcoming 'Astra' model family solved the ten major open problems previously announced, covering fields such as quantum complexity and lattice cryptography. The Astra model, which is being developed for long-running workloads and agent collaboration, formalized its proofs as Lean certificates and produced research texts that have now been publicly released. Additionally, reports indicate that the Astra model family may have disproved the Erdős unit distance conjecture in May, a problem that had been unsolved for approximately 80 years.
Researcher Noam Brown revealed that an internal version of the upcoming 'Astra' model family achieved these results, which included disproving the Connes’s rigidity conjecture and resolving several problems posed by Paul Erdős. Astra, which is designed for long-running workloads, formalized its arguments as Lean certificates for mathematical verification. While the release date remains unknown, the model’s achievements reportedly include resolving problems that had seen no progress for at least a decade.
OpenAI researcher Noam Brown revealed that an internal version of the upcoming 'Astra' model family achieved these results, solving 10 major open problems in fields such as quantum complexity and theoretical computer science. While the exact release name and date for Astra remain unknown, the model is designed for long-running workloads and allows AI agents to collaborate on complex tasks. Furthermore, Astra formalized its findings as Lean certificates to allow for automated mathematical verification.
