# AA-Robotics v0.1 benchmarks GPT-6 Astra for robot arm tasks

Digest AI · Robotics & Physical AI · published 2026-10-15T14:46:17Z

Canonical: https://digestai.news/story/aa-robotics-v0-1-benchmarks-gpt-6-astra-for-robot-arm-tasks

## Summary

A research team released **AA-Robotics v0.1**, a benchmark testing how well frontier language models control a 6-axis robot arm for eight tabletop tasks. The models—running in coding agent harnesses—must grasp objects, pour liquids, write text, and measure dimensions, judged on speed, cost, and accuracy. **GPT-6 Astra** led with a score of 69, but all models were **100 times slower** than humans, averaging 30 minutes per task versus 20 seconds for a person. Costs ranged from **$1.93 to $16.36 per task**, with input token costs dominating expenses (70% of the total). The team notes models often over-refine attempts after passing, suggesting gaps in physical reasoning. While unsuitable for production, the authors argue tracking such progress is critical as embodied AI evolves.

## Key points

- GPT-6 Astra scored highest at 69 on AA-Robotics v0.1’s eight tabletop tasks, but all models were 100x slower than humans
- Costs per task varied from $1.93 (GPT-6.1 Sol) to $16.36 (GPT-6 Astra), with input tokens accounting for 70% of expenses
- Models struggled with physical reasoning, often failing to recognize task completion or making irrecoverable mistakes

## Why it matters

This benchmark highlights the gap between AI’s theoretical capabilities and real-world physical dexterity, signaling progress but also limitations. For robotics, it underscores the need for better embodied AI—critical as automation expands beyond digital tasks. The cost and speed trade-offs also raise questions about practical deployment timelines.

## Sources

1. [Welcome to the era of embodied AI: Introducing AA-Robotics v0.1](https://artificialanalysis.ai/articles/aa-robotics) (artificialanalysis.ai, 2026-10-15)

Part of the developing story: [OpenAI's GPT-6 Expansion and Pricing Overhaul](https://digestai.news/thread/comparing-openai-models-on-aws-for-workload-efficiency) (10 stories)

## Cite

Digest AI, "AA-Robotics v0.1 benchmarks GPT-6 Astra for robot arm tasks", 15 October 2026, https://digestai.news/story/aa-robotics-v0-1-benchmarks-gpt-6-astra-for-robot-arm-tasks

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/aa-robotics-v0-1-benchmarks-gpt-6-astra-for-robot-arm-tasks.json