News — BenchCAD
benchcad.com · 431 words · saved by 1 readers
Execution-grounded benchmark for LLMs & multimodal models on programmatic CAD (CadQuery).
2026-07-09 OpenAI evaluated its GPT-5.6 family on BenchCAD OpenAI's GPT-5.6 launch post lists BenchCAD in its computer-use table, next to OSWorld and BrowseComp — the second frontier lab to report on the benchmark, and the first to run it across a whole model family. As published: GPT-5.6 Sol scores 70.6% without tools and 83.4% with a Python tool , with Terra at 62.3% / 78.2%, Luna at 63.1% / 73.9%, and GPT-5.5 at 44.4% / 55.8%. Scores exactly as printed in OpenAI's launch table (computer-use section, 2026-07-09). None are re-graded by BenchCAD. The Claude bars are Anthropic's own published v
saved by
related reading
- BigFinanceBenchbigfinancebench.com
- Robocurve — Real-world evaluations of physical AIrobocurve.org
- Summary of METR's predeployment evaluation of GPT-5.6 Solmetr.org
- GPT-5.6 Preview System Carddeploymentsafety.openai.com
- Noam Brown on X: "Implications of Large-Scale Test-Time Compute" / Xx.com
- Latency Scaling Differences for GPT and Claude Modelsepoch.ai
- GPT-5.6 Preview System Card - OpenAI Deployment Safety Hubdeploymentsafety.openai.com
- AINews | AINewsnews.smol.ai
- CAIS AI Dashboarddashboard.safe.ai
- TERMINAL-BENCHtbench.ai
- Two different tricks for fast LLM inferenceseangoedecke.com
- BenchmarkList: Track the Frontier of AI Capabilitiesbenchmarklist.com