Astra can do 34 consecutive simple math operations in latent space, 4x more than Sol
一项最新基准测试显示，GPT-6 Astra 能够在潜在空间中连续执行 34 次简单的数学运算，且无需使用思维链，这一表现远超其他模型。与 Sol 仅能完成 8 次运算相比，Astra 的能力是其四倍。即使是排名第二的 Claude Opus 4.6，也只达到了 12 步。这标志着大型语言模型在长链推理能力方面取得了显著的进步。
时间与来源
时间显示为 UTC
显示时区：UTC
本地时区尚不可用，暂时显示 UTC。
发布当时偏移：UTC+02026年9月9日 20:24 UTC
收录当时偏移：UTC+02026年9月9日 23:00 UTC
- 发布
- 2026年9月9日 20:24
- 收录
- 2026年9月9日 23:00
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 正常
档位是按信源手工设定的编辑判断，不是逐条打分。
After hearing rumors that GPT-6 Astra uses some form of recurrent depth, I decided to make a benchmark that specifically forces LLMs to do long chains of reasoning in latent space, without using chain-of-thought. The results for Astra are crazy: it can do 34 consecutive simple math operations (add, subtract, multiply and divide, all with small numbers) without using chain-of-thought! For comparison, Sol can only do 8!
I tested 24 other models (also from Anthropic) and none got anywhere close to Astra's result. What's even stranger is that there has been barely any progress on this benchmark (the second best result was Claude Opus 4.6, with 12 steps), until the release of Astra.
You can see the full results here, along with some speculation about what might be happening inside Astra that could explain this result:
https://maartenbaert.github.io/LatentMathBench/