AI modelsSeptember 26, 2026
17
š±
GPT-6 Astra Masters IKEA Assembly
GPT-6 Astra identifies IKEA shelf assembly errors with 80 percent accuracy.
#OpenAI#GPT-6#Visual Reasoning#Benchmark#Multimodal

š„ What happened
OpenAI's GPT-6 Astra hit 80% accuracy on the Furniture Assembly Benchmark (FAB), correctly spotting intentional mistakes in 60 photos of three IKEA builds. The previous best was 28% by Claude Opus 4.5 in November 2025. Astra also runs fastest, at a median 3 minutes per photo.
š” Why it matters
In ten months, error detection nearly tripled. That's the difference between a useless and a usable assistant for real-world repair and assembly. Astra is 2ā10x faster than prior leaders, making real-time guidance plausible for the first time.
ā” Our take
If you dismissed AI repair help in 2025, you were right. Now you're not. The real question: can OpenAI hold this lead, or will open-weight models close the seven-month gap?
The title, summary and analysis of this item were produced automatically by an AI system and have not been editorially reviewed. They may contain errors, bias or omissions ā when in doubt, read the linked original source.
Deep Dives & Similar Intelligence
ki-daily.
GPT-6 Astra Cracks 85-Year-Old Enigma
ResearchSeptember 22, 2026
ki-daily.
OpenAI Slashes GPT-6 Prices by Half
AI modelsSeptember 22, 2026

Robot AIs Execute Harmful Commands
Ethics & SecuritySeptember 19, 2026
SWE-Bench Pro V2: The Cheating Crackdown
Tools & ProjectsSeptember 23, 2026
ki-daily.
OpenAI's New Mental Health Benchmark
Ethics & SecuritySeptember 24, 2026