CADBench: 10 models tested on 105 Fusion 360 tasks, best pass rate only 24.6%
DynamicWebPaige · x · 2026-08-22
Seldon released CADBench, evaluating 10 models on 105 realistic CAD tasks in Fusion 360. Fable 5 and Gemini 3.7 Flash tie for #1 with only a 24.6% pass rate.
The notable signal: the co-leader is Gemini 3.7 Flash, not a frontier-tier model — the gap between "best" and "cheapest-good-enough" for computer-use CAD is already near zero, changing who gets to build in Fusion 360. A benchmark worth watching for how close agents are to building physical things.
Related event: CADBench: Gemini 3.7 Flash Ties for First, But Top Pass Rate Only 24.6%(3 posts)→
More from Research
- Five pharma giants federally train OpenFold3 on 20,000 private structures, +11 pts — jiqizhixin · 2026-10-16
- SWAN: a semantic-level text watermark that survives paraphrasing — burkov · 2026-10-16
- Interp researchers push back: 'networks are already clean' claim rests on dropout artifact — aryaman2020 · 2026-10-16
- LayerTrim: training-free pruning keeps reasoning LLMs intact on AIME — SeoulNatlUniv · 2026-10-16
- Simular presents neuro-symbolic computer use agents at BayLearn — xwang_lk · 2026-10-16
- Google's Hartmut Neven disputes chip test, outlines plan to link a brain to a quantum computer — Chris_Armstrong · 2026-10-16