GPT-6 Astra hits 80% on Epoch AI IKEA assembly benchmark
OpenAI's GPT-6 Astra reportedly reached eighty percent accuracy on Epoch AI's Furniture Assembly Benchmark, which evaluates whether a model can spot mistakes in deliberately mis-assembled IKEA furniture from photographs. The Decoder presents that result as a steep climb from the twenty-eight percent achieved by the best model in November 2025, framing a concrete visual error-detection task rather than generic captioning. Epoch AI is also cited as saying inference still takes about three minutes per photo, so real-time assembly guidance remains out of reach even as headline accuracy improves. Readers should treat the eighty percent figure as reported through one outlet, without independent verification in this packet, and without published detail on which IKEA items were tested, how accuracy was scored, or which failure modes persist at scale.
GPT-6 Astra hits 80% on Epoch AI IKEA assembly benchmark
The Decoder reports Epoch AI's Furniture Assembly Benchmark scores OpenAI's GPT-6 Astra at 80% accuracy on photos of deliberately mis-assembled IKEA pieces, up from 28% for the best model in November 2025. Epoch AI notes inference still takes about three minutes per photo, so it is not yet fast enough for real-time assembly guidance.
Key takeaway
The eighty percent IKEA assembly score is a reported Epoch AI benchmark result, not proof of a shipped, real-time consumer feature.
What happened
The Decoder reports that Epoch AI's Furniture Assembly Benchmark scores OpenAI's GPT-6 Astra at eighty percent accuracy on photos of deliberately mis-assembled IKEA pieces, up from twenty-eight percent for the best model in November 2025.
According to reporting summarized in the packet, GPT-6 Astra can analyze a photo and indicate whether an IKEA piece was assembled incorrectly, while Epoch AI notes inference still takes about three minutes per photo, which is not fast enough for real-time assembly guidance.
Evidence
GPT-6 Astra scored 80% on Epoch AI's Furniture Assembly Benchmark on photos of deliberately mis-assembled IKEA pieces.
The Decoder · attributed
The Decoder reports Epoch AI's Furniture Assembly Benchmark scores OpenAI's GPT-6 Astra at 80% accuracy on photos of deliberately mis-assembled IKEA pieces
The best model in November 2025 reached 28% on the same benchmark context.
The Decoder · attributed
up from 28% for the best model in November 2025
Epoch AI notes inference takes about three minutes per photo, insufficient for real-time guidance.
The Decoder · attributed
Epoch AI notes inference still takes about three minutes per photo, so it is not yet fast enough for real-time assembly guidance
Why it matters
If the benchmark trend holds, frontier multimodal models are moving toward practical visual defect localization, but multi-minute latency still blocks interactive assembly and field-use workflows.
Limits and uncertainties
The packet relies on The Decoder's reporting without independent verification of the eighty percent figure or benchmark methodology.
Available excerpts do not specify the IKEA items tested, how accuracy was measured, or dominant failure modes.
Reporting does not establish whether GPT-6 Astra capability is a shipped product feature or a research benchmark result.
Practical implications
Do not plan production real-time visual assembly coaching on these latency figures without separate latency and reliability testing.
Evaluate visual error-detection models on domain-specific mis-assembly photos rather than assuming headline benchmark scores transfer to your SKUs.
What to watch
Epoch AI publication or update of Furniture Assembly Benchmark methodology and item coverage.
Reported changes in per-photo inference time toward sub-minute interactive use.
Independent replication or additional publishers confirming the eighty percent score and failure-mode breakdown.
