SPACEXAI
4/5 workers samplingPOSSIBLE_MODEL_CHANGE · gpt-5
public · science

CAD Science Lab

Constraint CAD + scientific simulator tasks with measurable geometry.

Items
64
Revision
2026.04
Contamination
clean
Scorer
scorer.cad.v1
Dataset revision
[email protected]
Scorer
sim-oracle · scorer.cad.v1
License
GBU public suite
Kind
Public reproduction suite
ModelScore$/taskMedian sToolsEvidence
o4-pro81.4$0.63837.20.2LIVE_MEASURED
Grok 4 Heavy76.8$0.13221.90.2LIVE_MEASURED
Grok 471.2$0.03810.00.2LIVE_MEASURED
Gemini 2.5 Pro70.8$0.0258.40.2LIVE_MEASURED
GPT-569.4$0.06312.40.2LIVE_MEASURED
Claude 4.1 Opus68.1$0.26716.20.2LIVE_MEASURED
DeepSeek V3.166.4$0.00427.90.2REPRODUCED
Qwen3-Max63.8$0.00569.30.2LIVE_MEASURED
Claude 4.1 Sonnet62.4$0.04711.00.2LIVE_MEASURED
Mistral Large 358.2$0.0159.50.2LIVE_MEASURED
Llama 4 405B54.2$0.00438.70.2REPRODUCED
GPT-5 Mini52.1$0.00314.80.2LIVE_MEASURED
Gemini 2.5 Flash51.6$0.00113.20.2LIVE_MEASURED
Grok 3 Mini48.6$0.00163.80.2LIVE_MEASURED
Llama 4 70B42.8$0.000941.80.2REPRODUCED