#推理

GPT-6 Astra 现已面向 Work/Codex 中的所有 Pro、Enterprise 和 Business Premium 用户开放，并已在 API 中提供。 我们接下来将开始向 Plus 和 Business 用户推送。 感谢大家的耐心等待。

Checking that a major mathematical proof is correct can take years. Formalization—converting the mathematical reasoning into a form computer proof assistants like Lean can verify—can help. Last month, Claude completed the first formalized proof of Fermat’s Last Theorem, one of the most famous theorems of all time. This was a project experts thought would take many years. It is the largest Lean proof ever written. Fermat’s Last Theorem was first proven in 1995 by Sir Andrew Wiles, more than 350 years after it was conjectured. Our proof, which totals over 13 million lines of code, provides machine verification. More importantly, it proves over 29,000 other theorems that the proof requires, across many areas of math which had never before been formalized. We see this as a major step in the long process of firming up the core of mathematical knowledge, building on work from three centuries of mathematicians and hundreds of contributors to Lean and Mathlib. We are optimistic that AI-assisted verification of mathematical proofs will help reduce the burden of refereeing mathematics in an era where more proofs are being produced than ever before. You can read about the process on our Science Blog: https://www.anthropic.com/research/formalizing-fermats-last-theorem And see the complete proof on GitHub: https://github.com/anthropics/fermats-last-theorem

Our evaluation results for Orbis 1.0: Across automated benchmarks, Orbis scores highest on DOVER aesthetic and technical quality, and on VideoAlign visual and motion quality. It also leads three physical-plausibility protocols: VideoPhy-2, Physics-IQ, and VBench-2.0 Physics. In a randomized human Arena study of state-of-the-art real-time interactive video systems, Orbis ranked first in overall preference and first in temporal stability, the prerequisite for long-form generation. Full report: https://www.visko.ai/blog/papers/files/Visko_Orbis_1_0__A_Live_Model_for_Real_TimeInteractive_Long_Video_Generation.pdf

GPT-6 Astra is on the pareto-frontier of cost efficiency due to being EXTREMELY token efficient. It is in a whole league of it's own. It is cheaper than Gemini 3.8 Flash per task (a model 13x cheaper than Astra).

GPT-6 Astra has set a new ECI record, with a score of 169. This is a substantial jump from the prior best (163), but is within our uncertainty range for the reasoning-era ECI trend. Astra also set new records on our math, continual learning, and game-puzzles benchmarks. On our long-horizon coding benchmark, MirrorCode, Astra ranks between Opus 4.7 and Fable 5. OpenAI gave us pre-release access to test Astra. Charts and more details for Astra’s individual benchmark results in the thread.

由 @OpenAI 推出的 GPT-6 Astra 在 ARC-AGI 上达到 SOTA： - Astra 在 ARC-AGI-3 上得分 63%，通过新的提供商适配器测试框架可达 99% - 它在 96% 的 ARC-AGI-3 关卡上超越了人类表现 - 它构建了我们所见过的、对新环境最精确的符号模型 我们的分析：

由 @OpenAI 推出的 GPT-6 Astra 在 ARC-AGI 上达到 SOTA： - Astra 在 ARC-AGI-3 上得分 63%，通过新的提供商适配器测试框架可达 99% - 它在 96% 的 ARC-AGI-3 关卡上超越了人类表现 - 它构建了我们所见过的、对新环境最精确的符号模型 我们的分析：

Luna is the first closed source model in a long time to top the OpenRouter charts! Great job to the OpenAI team! Incredible combination of price and performance.

我们在 WANDR 上评估了 GPT-6 Astra。它以每任务 $11.98 的成本得分 0.682，是我们测试过的所有模型中的最高分。 GPT-6-Astra 比 Fable 5.1 高 13.5%，成本低 6.1%；比 Opus 5 高 27.0%，成本高 3.3%。

OpenAI 的发布视频越来越出色了。

当我们发布 ARC 3 时，有人问我：“你觉得前沿模型什么时候能把它刷满？”我回答说：“大约一年吧，不过这取决于它被专门针对的程度。” 那是 6 个月前的事了，所以 Astra 所代表的进展大约比我预想的快了 2 倍。我认为进展的速度会让很多人吃惊，而新模型能做到的事，将会挑战人们通过使用前几代模型所形成的对 AI 的看法。

New OpenAI repo with a Lean formalization by GPT-6-Astra proves that there are infinitely many pairs of consecutive primes whose distance is at most 186 https://github.com/openai/PrimeGaps186

We invite you to read the full TRACES technical report Inside: the definitions, evaluation rubric, and registry of problems spanning biomedicine, clinical translation, and frontier-model engineering. Led by Apodex Lead Scientist @profshengwang, the report brings together the technical framework underpinning this effort. Behind it: 10 STEM PhDs, two months of research, and 561 industries surveyed to identify 423 problems worth solving—real, high-stakes questions that can take human experts years to work through. Bring a problem. Bring a solver. Both doors are open. *Submit a Problem: https://traces.apodex.com/submit-your-problem *Submit a Solver: https://traces.apodex.com/submit-your-solver