Benchmark
BenchCAD
Can a model reconstruct parts, assemblies and circuit boards from engineering drawings and rendered images?
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
The 2.1 Core evaluation uses the same 100 cases as 2.0. Models write code, inspect the rendered result and revise their submissions. The benchmark awards partial credit.
Methodology
The standard 2.1 harness gives each case up to 60 rounds and a shared output-token budget. The project reports customized tool harnesses separately. The scoring and budgets changed between 2.0 and 2.1, so scores across those releases cannot be compared directly.