AA-3D benchmark launches: Claude leads as four models independently embed Sagrada Família magic square
ArtificialAnlys · x · 2026-10-22
Artificial Analysis launched AA-3D, a benchmark measuring how well models build interactive 3D scenes: each model acts as an agent in an offline sandbox, turning a written brief into a three.js page, ranked by blind pairwise Elo votes.
v1.0 leaderboard
- Claude Sonnet 5.5 tops at Elo 1194, but the top four (all Anthropic and OpenAI models) are statistically tied
- Meta's Muse Spark 1.3 is seventh (1035)
- Best open weights models Kimi K3 (982) and MiMo-V2.6-Pro (977) rank 11th and 12th, 200+ points behind
Fun finding: all four models spontaneously added the Sagrada Família magic square (sum 33) to their scenes, offline and unprompted—Claude Fable 5.1 and Sonnet 5.5 displayed it visibly, while Grok 4.7 and Step 5 Preview hid it behind a pillar.
More from Models
- Video Arena Elo: Wan 3.0 Tops at 1173, Kling 3.0 Slumps to 17th — ArtificialAnlys · 2026-10-22
- Artificial Analysis Video Arena Turns Two: From 8 Models to 140+ Benchmarked — ArtificialAnlys · 2026-10-22
- Anthropic quietly publishes 'Introducing Claude Duet' article on its site — Tiny-Examination5627 · 2026-10-22
- OpenAI Brings Frontier Models and Codex Agentic Workflows to Google Cloud — OpenAI · 2026-10-22
- LangSmith Fine-Tuning Turns Agent Traces Into Models: 9B Beats Frontier on Task Benchmark — Hacubu · 2026-10-22
- AI agents power growing wave of cyberattacks, with successful reaches costing just $50 — fortune · 2026-10-22