OpenAI’s latest model, GPT-6 Astra, has demonstrated the ability to identify assembly errors in IKEA furniture from photographs. According to reports from The Decoder, the model has achieved an 80 percent accuracy rate in detecting where a user may have made a mistake during the construction process. This represents a significant leap from the performance of AI models recorded in late 2025.
Significance of Visual Reasoning
In November 2025, the highest-performing model recorded an accuracy rate of just 28 percent for similar furniture assembly tasks. The current jump to 80 percent indicates a rapid improvement in spatial reasoning and visual analysis within large multimodal models. Epoch AI noted that while the accuracy has improved significantly, the processing speed is currently insufficient for real-time guidance during the assembly process. However, the organization suggests the performance gap is narrowing. This development highlights a shift toward AI that can interact with the physical world through visual inputs, moving beyond text-based or simple image-recognition tasks.
Unclear Parameters and Limitations
The announcement did not specify the exact hardware requirements for running GPT-6 Astra or whether this capability will be available to the general public in a standalone application. It is also unknown which specific IKEA furniture pieces were used in the testing or if the model’s accuracy varies depending on the complexity of the item. Furthermore, the reporting did not clarify if the model can provide step-by-step corrective instructions or if it simply identifies that an error has occurred. Future updates may focus on reducing latency to allow for a live camera feed to assist users as they work.
As spatial intelligence matures, GPT-6 Astra represents a step toward AI assistants that can troubleshoot physical tasks. The speed at which the model processes visual data remains the primary hurdle for practical, real-time application.
Lead image: Alfredo Marco Pradil / Pexels
