GPT-6 Astra on robot arms

GPT-6 Astra on robotic manipulation

GPT‑6 Astra on robotic manipulation September 4, 2026. A follow‑up to our comparison of Claude Fable 5 and Fable 5.1. We gave OpenAI’s GPT‑6 Astra control of the same YAM arms under the same Inspect Robots agent policy, on the same two tasks: “Pick up the red block from the table and place it inside the bowl.” “Pick up the round blue puzzle piece by the knob at its center and place it into the matching circular groove in the board.”

GPT-6 Astra 在机器人操作上的表现 2026 年 9 月 4 日。这是我们对比 Claude Fable 5 和 Fable 5.1 后的后续报道。我们让 OpenAI 的 GPT-6 Astra 在相同的 Inspect Robots 代理策略下控制相同的 YAM 机械臂,并执行相同的两项任务:“从桌子上拿起红色积木并放入碗中”以及“通过中心旋钮拿起圆形的蓝色拼图块,并将其放入板上匹配的圆形凹槽中。”

On the bowl task Astra placed the block in 19 of 20 trials, against Fable 5.1’s 8 of 20 and Fable 5 in 1 of 20, in 2.5 minutes per trial to Fable 5.1’s 6.8, at an estimated $0.94 per run to $2.12. The puzzle task is a different story: Astra completed the insertion 2 times in 20 against Fable 5.1’s 2 in 20. It reaches the groove and stalls at the same final step Fable does, at $1.36 per run to $2.18.

在碗任务中,Astra 在 20 次试验中成功放入积木 19 次,相比之下 Fable 5.1 为 8 次,Fable 5 为 1 次;平均每次试验耗时 2.5 分钟,而 Fable 5.1 为 6.8 分钟;预估单次运行成本为 0.94 美元,而 Fable 5.1 为 2.12 美元。拼图任务的情况则有所不同:Astra 在 20 次试验中完成了 2 次插入,与 Fable 5.1 的 2 次持平。它能到达凹槽位置,但在与 Fable 相同的最后一步停滞,单次运行成本为 1.36 美元,而 Fable 5.1 为 2.18 美元。

Scoring

Every trial was scored by a human grader on the highest stage it reached, so a run that fails still records how far it got. The rubric is unchanged from the Fable report: 0: No purposeful approach 1: Made contact with the object 2: Lifted the object clear of the table 3: Positioned it above the deposit point 4: Placed it in its final position

评分标准

每次试验均由人工评估员根据其达到的最高阶段进行评分,因此即使失败的运行也会记录其进展程度。评分细则与 Fable 报告保持一致: 0:没有目的性的接近 1:与物体接触 2:将物体从桌面上提起 3:将其定位在放置点上方 4:将其放置在最终位置

Results Summary

TaskModelMean stageCompletions RateEst. cost/runMinutes/run
Block into bowlFable 51.305%$2.698.2
Block into bowlFable 5.12.4040%$2.126.8
Block into bowlGPT-6 Astra3.9595%$0.942.5
Puzzle into grooveFable 51.500%$2.637.9
Puzzle into grooveFable 5.12.3510%$2.185.9
Puzzle into grooveGPT-6 Astra2.0010%$1.363.4

结果摘要

任务模型平均阶段完成率预估单次成本单次耗时(分钟)
积木入碗Fable 51.305%$2.698.2
积木入碗Fable 5.12.4040%$2.126.8
积木入碗GPT-6 Astra3.9595%$0.942.5
拼图入槽Fable 51.500%$2.637.9
拼图入槽Fable 5.12.3510%$2.185.9
拼图入槽GPT-6 Astra2.0010%$1.363.4