Kimi K3.1 Pro scores 12/46 on IKEA 3D assembly test
Five AI models attempted to turn a 24-page IKEA ALEX manual into an interactive 3D build. GPT-6.1 Sol via Codex and Claude Opus 5.5 tied at 46/46 on the manual rubric, while Kimi K3.1 Pro scored just 12/46 for assembly correctness despite a working interface.
- Sol via Codex finished in 19 min 18 s, Opus took 40 min 55 s
- Kimi scored 10.5/11 on features but 12/46 on correctness
- Astra lost a point for a transposed screw ID: 101907 instead of 110907
- Gemini 4 Argon's run cost an estimated $2.5562 at list price
Read next
AI