Today for AI
HOT RADAR
llmHEAT 23.1°

AI CURATED HOT STORY · 10/13/2026

Artificial Analysis Launches AA-3D Benchmark: Evaluating LLMs' Interactive three.js Scene Construction via Elo Rankings, with Claude Sonnet 5.5 Leading

3 reports archived1 independent sourcesupdated 10/14/2026, 08:44:35
Synthesis & Latest Updates
1 Sources Cross-Validated

Artificial Analysis has launched the AA-3D benchmark to evaluate large language models' ability to construct interactive three.js scenes from text briefs within an offline sandbox. Moving beyond static code checks, it employs an Elo ranking system derived from over 10,000 human blind votes, directly reflecting user preferences for visual quality and interaction.

Technically, the benchmark allows models to iteratively build web pages using Blender, Rapier physics engine, and Python. Claude Sonnet 5.5 leads with a score of 1194, statistically tied with several other Anthropic and OpenAI models. Notably, performance does not correlate linearly with cost; while GPT-6 Luna is the cheapest, it ranks eighth, highlighting that top-tier models still hold significant advantages in complex spatial logic and rendering fidelity.

This release provides developers with a more realistic evaluation metric, filling gaps in assessing dynamic interaction and visual quality. It encourages the industry to reconsider AI's potential in embodied intelligence and frontend automation, pushing models toward comprehensive agents capable of spatial understanding and real-time rendering, with profound implications for game development, digital twins, and Web 3D ecosystems.

LATEST/Current rankings show narrowed gaps among top models, yet low-cost options like GPT-6 Luna struggle with complex scenes. Future focus lies in optimizing physical engine call stability and spatial logic consistency during multi-turn iterations to enhance high-fidelity 3D content generation efficiency.

HEAT TRENDHourly heat curve

36 fully observed hours
Now
23.1
Peak
40.309:00
24h change
-26%

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

TIMELINECoverage timeline

Total 3 reports · Latest first
  1. Artificial Analysis (@ArtificialAnlys)T1·72 pts
    • Claude Sonnet 5.5 leads the Coding Agent Index with 68 points in max effort mode, while xhigh mode achieves 63 points with superior cost-efficiency.
    • The xhigh mode costs only $3.33 per task compared to $14.19 for max mode, reducing costs by ~77% for a marginal loss of 5.5 index points.
    • Max mode improves performance by generating 4.8× more output tokens and taking 3.4× more turns, boosting Terminal-Bench 4.0 scores from 58.1% to 66.2%.
  2. Artificial Analysis (@ArtificialAnlys)T1·78 pts
    • AA-3D uses an 'agent + sandbox' approach, allowing models to iterate on 3D web pages using tools like Blender, Rapier physics, and Python.
    • Scoring relies on large-scale blind human voting (Elo ranking), ensuring evaluation of final visual and interactive quality rather than just static code analysis.
    • Data shows performance is not linearly correlated with cost; the cheapest model, GPT-6 Luna, outperforms many expensive ones, though top-tier models still lead in complex scene construction.
  3. Artificial AnalysisT1·68 pts
    • AA-3D operates by having models act as agents in an offline sandbox to generate three.js code from text briefs.
    • The scoring system replaces static metrics with human expert blind pairwise voting converted into Elo rankings, reflecting real user preferences.
    • Claude Sonnet 5.5 currently tops the leaderboard, though the gap between leading Anthropic and OpenAI models is statistically insignificant.