Artificial Intelligence
OpenAI model scores 80% on furniture benchmark
Epoch AI said OpenAI's GPT-6 Astra reached 80% on a furniture-assembly photo benchmark using IKEA items, ahead of several Claude models and open-weight alternatives.
- Epoch AI's Furniture Assembly Benchmark used photos of three IKEA items assembled with deliberate errors and asked models to identify the mistakes from the instructions.
- The article said OpenAI's GPT-6 Astra scored 80% at three minutes per photo.
- It said Claude Opus 4.5 had scored 28% in November 2025, while Claude Fable 5.1 scored 70% and Claude Opus 5 scored 61%.
- The article said Chinese open-weight models, including Kimi K3, lagged the leaders by at least seven months.
- The researchers said the system was still too slow for real-time assembly help, though it might later assist with car repairs or appliance fixes.
Epoch AI's Furniture Assembly Benchmark showed models photographs of three IKEA pieces with planted mistakes. The models had to compare the images with the instructions, find the errors and explain them.
OpenAI's GPT-6 Astra was said to have scored 80% at three minutes per photo. The article said this placed it ahead of Claude Fable 5.1 at 70% and Claude Opus 5 at 61%.
For comparison, the article said Claude Opus 4.5 had scored 28% in November 2025. It also said Chinese open-weight models such as Kimi K3 were at least seven months behind the leaders.
The researchers said the pace was still too slow for live help while someone assembled furniture. They said the approach could one day help with car repairs or appliance fixes, and the article added that Astra also performed well on visual robotic tasks.
Named in this story
Companies
- OpenAI
- was the company behind GPT-6 Astra
- IKEA
- provided the furniture pieces used in the benchmark
Organisations
- Epoch AI
- created the Furniture Assembly Benchmark
Products and systems
- GPT-6 Astra
- scored 80% on the benchmark
- Furniture Assembly Benchmark (FAB)
- tested models on IKEA assembly photos with mistakes
- Claude Opus 4.5
- scored 28% in November 2025
- Claude Fable 5.1
- scored 70%
- Claude Opus 5
- scored 61%
- Kimi K3
- was among the open-weight models that lagged behind
How the source tells it
The piece was a short performance update with a mildly boosterish tone, using a sharp comparison of model scores and a headline built around novelty.
- novelty the headline and opening framed the capability as a new arrival
- boosterism the article emphasised a large score jump and singled out one model as especially strong