OpenAI's GPT-6 Astra can now identify misassembled IKEA furniture from photos with 80 percent accuracy, up sharply from 28 percent in November 2025. The model represents significant progress toward real-time assembly guidance, though speed improvements are still needed.
OpenAI's latest multimodal model, GPT-6 Astra, has demonstrated a marked improvement in identifying assembly errors in IKEA furniture. When shown photos of assembled pieces, the system correctly identifies mistakes 80 percent of the time—a massive leap from the 28 percent accuracy achieved by the best available model just months ago.
The capability works by analyzing images and pinpointing where assembly went wrong. This suggests potential applications beyond furniture troubleshooting, including quality control in manufacturing and step-by-step assembly verification for users.
However, the model still faces a critical limitation: speed. According to Epoch AI, current processing is not yet fast enough to provide real-time guidance during assembly. The lag would make it impractical for someone actively building furniture, who would need immediate feedback as they progress through each step.
Despite this constraint, the trajectory is encouraging. The dramatic accuracy gains in recent months suggest that processing speed improvements could arrive relatively soon. Real-time assembly assistance could eventually reduce frustration and returns for flat-pack furniture retailers and improve the customer experience.
The progress reflects broader advances in vision-language models. GPT-6 Astra's ability to understand spatial relationships, identify deviations from instructions, and communicate findings demonstrates the sophistication of current AI systems in handling visual reasoning tasks.
For IKEA and similar retailers, such technology could eventually power in-app guidance systems or customer service tools. Users experiencing assembly difficulties could photograph their progress and receive instant feedback on errors before proceeding further.
OpenAI has not announced plans to deploy this capability publicly, but the rapid improvement suggests it could be integrated into consumer-facing products within months rather than years.
Bloomberg journalists will host a live Q&A session Friday, September 25 at 9 a.m. EDT to discuss the escalating US-China competition in artificial intelligence development.
Two-thirds of IT leaders report measurable AI results, but only 8 out of 160 surveyed consider their achievements significant enough to warrant interrupting the CEO's vacation.
A study of over 3,000 participants found that access to AI drops willingness to admit ignorance from 44% to just 3%. Users felt confident despite being wrong twice as often as those without AI.
Anthropic's latest Claude model uses 95% fewer em dashes while producing longer responses overall. The shift suggests changes to how the AI generates text.