GPT-6 Astra, Looped Transformers, and Hidden Reasoning

GPT-6 Astra、循环 Transformer 与隐藏推理

A Look at Recurrent Depth, Hidden Chains of Thought, and Recent Research on Looping Transformer Blocks 深入探讨循环深度、隐藏思维链以及关于循环 Transformer 模块的最新研究

Sebastian Raschka, PhD | Sep 09, 2026

A lot has happened in the last few weeks. I am sure that OpenAI’s GPT-6 Astra is top of mind for everyone right now. In particular, thoughts on its performance, the looped transformer/recurrent depth aspects, and rumors that Astra is “hiding” its reasoning trace (i.e., chain of thought). 过去几周发生了许多大事。我相信 OpenAI 的 GPT-6 Astra 目前是每个人关注的焦点。特别是关于其性能、循环 Transformer/循环深度方面的思考，以及关于 Astra 正在“隐藏”其推理轨迹（即思维链）的传言。

So, in this article, I want to start with some brief impressions of Astra and some thoughts on where all this is headed. Then, I will discuss, in detail, what “looped transformers” are, and how (or rather, if) this relates to hiding chains of thought. 因此，在本文中，我想先谈谈对 Astra 的初步印象以及对未来发展方向的一些思考。随后，我将详细讨论什么是“循环 Transformer”，以及它与隐藏思维链之间有何关联（或者说，是否真的存在关联）。

Lastly, after covering the basics of the looped transformer, I wanted to highlight some new insights from recent research papers on the topic. 最后，在介绍完循环 Transformer 的基础知识后，我想重点介绍近期相关研究论文中的一些新见解。

1. GPT-6 Astra impressions

1. GPT-6 Astra 初步印象

First things first. Before getting into the architecture rumors and related research literature, let me briefly summarize some GPT-6 Astra observations and tidbits. 首先，在深入探讨架构传言和相关研究文献之前，让我先简要总结一下对 GPT-6 Astra 的一些观察和心得。

Last week, OpenAI’s new GPT-6 Astra was released with a big fanfare. I used it over the past couple of days, and it’s an exceptionally good model, likely the best I’ve used as of this writing. But what, exactly, has it improved, and how? 上周，OpenAI 的新模型 GPT-6 Astra 在万众瞩目中发布。过去几天我一直在使用它，这是一个非常出色的模型，很可能是截至目前我所用过的最强模型。但它究竟改进了什么，又是如何改进的呢？

1.1 Astra benchmarks

1.1 Astra 基准测试

Astra is the best model I’ve used so far, and it’s disproportionately good at 3D rendering and animation tasks (relative to other models). With that, I mean that while it leapfrogs its GPT-5.6 predecessor in practically all categories (writing, math, coding, and more), it especially does so when it comes to graphical demos. Astra 是我迄今为止用过最好的模型，它在 3D 渲染和动画任务上的表现（相对于其他模型）好得不成比例。我的意思是，虽然它在几乎所有类别（写作、数学、编程等）上都超越了其前身 GPT-5.6，但在图形演示方面表现尤为突出。

We can see this also reflected in the benchmarks. For instance, GPT-6 Astra is really good at math and coding, as shown below. 这一点也反映在基准测试中。例如，如下图所示，GPT-6 Astra 在数学和编程方面表现非常出色。

(Figure 1: Selection of three popular coding benchmarks and one challenging math benchmark. More benchmarks are shared on the Astra release blog: https://openai.com/index/gpt-6-astra/) （图 1：三个热门编程基准测试和一个具有挑战性的数学基准测试的选择。更多基准测试可在 Astra 发布博客上查看：https://openai.com/index/gpt-6-astra/）

One of the highlights (not shown in the figure) is that Astra also achieves 99.9% on the ARC-AGI-3 benchmark (GPT-5.6 Sol only 7.8%), which measures a mix of solving logic puzzles and generalization. However, the math, coding, and computer use benchmarks are more interesting because they are closer to real-world use. 亮点之一（图中未显示）是 Astra 在 ARC-AGI-3 基准测试中达到了 99.9% 的准确率（GPT-5.6 Sol 仅为 7.8%），该测试衡量的是逻辑谜题解决能力与泛化能力的综合水平。然而，数学、编程和计算机使用基准测试更有趣，因为它们更接近现实世界的应用。

Coming back to the Artificial Analysis Coding Agent Index v1.4 (lower right in the previous figure), which blends several agentic coding tasks, GPT-6 Astra is clearly at the frontier, but it doesn’t pull ahead by leaps and bounds. This can also be seen in the general Artificial Analysis Intelligence Index shown below, which blends different types of tasks, not just coding tasks. 回到 Artificial Analysis Coding Agent Index v1.4（上图右下角），它融合了多种智能体编程任务，GPT-6 Astra 显然处于前沿，但并没有实现跨越式的领先。这一点也可以从下方的通用 Artificial Analysis Intelligence Index 中看出，该指数融合了不同类型的任务，而不仅仅是编程任务。

(Figure 2: Artificial Analysis Intelligence Index via https://artificialanalysis.ai/#intelligence) （图 2：Artificial Analysis Intelligence Index，来源：https://artificialanalysis.ai/#intelligence）

Now, the big advantage of Artificial Analysis benchmarks is that they are independent and thus may be a bit more trustworthy than self-evaluated benchmarks by model developers. 目前，Artificial Analysis 基准测试的一大优势在于它们是独立的，因此可能比模型开发者自评的基准测试更值得信赖。

The harness setup depends on the benchmark. For example, GDPval-AA and AA-Briefcase use their open-source, minimal Stirrup harness across the different LLMs they compare. In the Intelligence Index v4.2 shown above, Terminal-Bench v2.1 uses Terminus 2, and τ³-Banking uses the τ-Bench harness. The separate Coding Agent Index also compares different coding-agent harnesses. 测试框架的设置取决于基准测试本身。例如，GDPval-AA 和 AA-Briefcase 在比较不同的 LLM 时，使用了它们开源的、极简的 Stirrup 框架。在上述 Intelligence Index v4.2 中，Terminal-Bench v2.1 使用了 Terminus 2，而 τ³-Banking 使用了 τ-Bench 框架。独立的 Coding Agent Index 也比较了不同的编程智能体框架。

For evaluations that use a shared harness, this makes it more of an apples-to-apples comparison. At the same time, during model training, models are typically developed with one primary harness in mind (and fine-tuned less on other harnesses). Plus, the primary harness is often developed to suit and amplify a model’s strengths. 对于使用共享框架的评估，这使得比较更加公平。与此同时，在模型训练过程中，模型通常是针对一个主要框架进行开发的（而在其他框架上的微调较少）。此外，主要框架通常是为了适应并放大模型的优势而开发的。

So, some of the agentic evaluations might underestimate how well Astra performs in its primary harness. How much this affects its Intelligence Index score would need to be tested by comparing Astra across harnesses on the same tasks. 因此，一些智能体评估可能会低估 Astra 在其主要框架中的表现。这在多大程度上影响了其 Intelligence Index 得分，还需要通过在相同任务上跨框架比较 Astra 来进行测试。

As a side note, as a colleague recently suggested to me (as also recommended by the Claude Code lead), it’s maybe not a bad idea to delete (/archive) some of your existing AGENTS.md contents and SKILL.md files, as newer LLMs have become more efficient at understanding the prompt and solving the problem at hand. The extra hand-holding could unnecessarily constrain newer models and lead to worse solutions. 顺便提一下，正如一位同事最近向我建议的那样（Claude Code 的负责人也曾推荐过），删除（或归档）一些现有的 AGENTS.md 内容和 SKILL.md 文件或许是个不错的主意，因为较新的 LLM 在理解提示词和解决当前问题方面已经变得更加高效。过多的“手把手”指导可能会不必要地限制新模型，并导致更差的解决方案。

Of course, I am not suggesting never using SKILL.md files again, but for some workflows, because they can improve efficiency upon reuse, since the model doesn’t have to rediscover them. But what I am suggesting is that some workflows don’t need describing, and “old” descriptions may no longer be ideal, and the LLM may be able to come up with better solutions. So, it’s perhaps time to update or regenerate said instruction files. 当然，我并不是建议永远不再使用 SKILL.md 文件，因为对于某些工作流，它们可以通过复用来提高效率，模型无需重新探索。但我建议的是，有些工作流并不需要描述，而且“旧”的描述可能不再理想，LLM 或许能提出更好的解决方案。因此，现在也许是更新或重新生成这些指令文件的时候了。

1.2 Computer use capabilities

1.2 计算机使用能力

GPT-6 Astra seems to be exceptionally strong in image and rendering tasks. When these tasks involve interacting with graphical user interfaces, they also demonstrate computer-use capabilities, meaning the model operates software on your local computer through the Codex/ChatGPT app. GPT-6 Astra 在图像和渲染任务上似乎异常强大。当这些任务涉及与图形用户界面交互时，它们也展示了计算机使用能力，这意味着模型可以通过 Codex/ChatGPT 应用程序操作你本地计算机上的软件。

Computer use is where the model really shines compared to others, and anything graphic-related also makes for interesting and intuitive demos on social media platforms. There are tons of examples of impressive demos out there, from modeling rendering New York City in blender to virtual open house tours. 与其它模型相比，计算机使用能力是该模型真正闪光的地方，任何与图形相关的内容在社交媒体平台上都能带来有趣且直观的演示。网上有大量令人印象深刻的演示示例，从在 Blender 中渲染纽约市模型到虚拟房屋参观。

To pick one example, below is a comparison where I had GPT-6 Astra Medium and High redraw a picture of me in a browser version of MS Paint using the mouse on my computer (not Extra High and Max, because I didn’t want to waste all my tokens :)). 举个例子，下面是一个对比，我让 GPT-6 Astra Medium 和 High 版本使用我电脑上的鼠标，在浏览器版的 MS Paint 中重绘了我的照片（没有使用 Extra High 和 Max 版本，因为我不想浪费所有的 Token :)）。

This highlights not only the model’s artistic capabilities but, more importantly, its ability to use tools on one’s computer (in this case, Paint; you can see the model using the interface via the mouse cursor). 这不仅突显了该模型的艺术能力，更重要的是，它展示了模型使用计算机工具的能力（在本例中是 Paint；你可以看到模型通过鼠标光标操作界面）。

This is not the first model that, inside a harness, is capable of general computer use. For example, I successfully used GPT models for some UI tasks (e.g., expense-related tasks in Excel) and so on since earlier this year. 这并不是第一个在框架内具备通用计算机使用能力的模型。例如，自今年年初以来，我已经成功地将 GPT 模型用于一些 UI 任务（例如 Excel 中的费用相关任务）等。