OpenAI's GPT-6 Astra can now tell you exactly where you screwed up your IKEA shelf
80 percent. That's GPT-6 Astra's accuracy on spotting bad IKEA assembly — up from 28 percent in November. Vision is closing the real-time gap.
Why it matters
GPT-6 Astra shows material progress in multimodal vision reasoning over a short interval, moving from parlor-trick accuracy to practical territory. The benchmark itself (IKEA assembly correctness) is niche, but the velocity and the closing gap to real-time latency matter to frontier watchers tracking capability density and inference speed.
The key facts
9 to knowGPT-6 Astra: 80% accuracy on IKEA assembly correctness detection
November 2025 baseline: 28% accuracy (same task)
Improvement: 52 percentage points in ~10 months
Limitation: speed not yet fast enough for real-time assembly guidance, but gap closing
Source: Epoch AI benchmark tracking
GPT-6 Astra achieves 80% accuracy on IKEA assembly correctness detection
Previous best model (November 2025) achieved 28% on same task
Speed remains insufficient for real-time assembly guidance per Epoch AI
Source: The Decoder; benchmark source: Epoch AI
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: OpenAI's GPT-6 Astra can look at a photo and tell whether an IKEA furniture piece was assembled incorrectly, hitting an 80 percent accuracy rate. Back in November 2025, the best model managed just 28 percent. According to Epoch AI, the speed isn't quite fast enough yet for real-time assembly…