OpenAI's GPT-6 Astra model can now identify assembly errors in IKEA furniture by analyzing photographs, reaching an 80% accuracy rate.

OpenAI's GPT-6 Astra model has demonstrated the ability to identify mistakes in IKEA furniture assembly by comparing photographs of the process against official instructions. The model achieved an 80% accuracy rate in the Furniture Assembly Benchmark (FAB) developed by Epoch AI. This represents a significant improvement over the 28% accuracy rate recorded by the top-performing model in November 2025.

Benchmark Performance and Competition

The Furniture Assembly Benchmark evaluates AI models by having them analyze photos of three IKEA furniture pieces that contain deliberate assembly errors. Following GPT-6 Astra's 80% performance, other models also showed progress: Claude Fable 5.1 achieved 70% accuracy, while Claude Opus 5 reached 61%. Chinese open-weight models, such as Kimi K3, currently trail these leading models by at least seven months in performance.

Future Applications and Limitations

While GPT-6 Astra currently processes images at a rate of three minutes per photo, researchers note that this speed is not yet sufficient for real-time assembly guidance. However, the technology is viewed as a potential tool for future applications in fields such as car repairs and appliance maintenance. The progress is considered notable given that AI models struggled with simpler visual tasks in the recent past.