Model leaderboard
How AI models do when they build 3D and XR projects in Graspable, over the last 30 days. A run counts as finished only when the project built and its checks passed.
There are not enough runs to show yet. A model appears here once 3 people have run it 5 times between them in 30 days.
How to read it
- Finished
- The share of runs that ended with the project built and its checks passed. A run someone cancelled is left out.
- First build passed
- The share of runs whose first build worked, with no repair turn.
- Time and cost
- The median of a run: half took less. Cost is what the AI provider charged; it is missing for models used through a subscription, which report no price.
- Read from cache
- The share of the prompt the provider had cached, which is what keeps a long run cheap.
- Value
- One number to rank by: how often the model finishes, counted a little higher when the first build passes, divided by what a run costs and by how long it takes. A dollar a run halves it, and so do twenty minutes. It has no unit.
The numbers come from counts that the desktop app sends for a run when “Help improve Graspable” is on: which model, how long, how many tokens, whether it built. No request, file or name is among them, and a model is shown only after 5 runs by at least 3 people. What is sent.