Hacker News
SketchAgentBench – evaluate LLMs on 3D spatial reasoning(sketchagent.dev)
9 points by wsxiaoys 1 day ago | 5 comments
[–] anilgulecha 1 day ago
It's only benchmarks where I hear about Google's models being frontier. Never really get to use it because they don't have a usable subscription model on par with Claude/Openai ones in terms of usage.
[–] dogma1138 18 hours ago
[–] anilgulecha 18 hours ago
These are not equal to the other 2 in usage in amount of inference, like I mentioned. Pain to get work with Pi as well.
[–] gorpy7 1 day ago
i had astra route some ducting thru joists and beams and other crap and i was so surprised and rejoicing because it knocked it out of the park! this is sketchup pro. i’m not especially good at sketchup. i also had it do some plumbing and again it was good. i wanted proper slope on all the joints and pipes and if i needed to adjust a route- boy was that aggravating! but then asked astra and just sipped my pop.
[–] vivzkestrel 1 day ago
- can we get a benchmark on game decompilation / figuring out?