[xiphias2]

Another project without running real benchmarks. It's very easy to generate tokens, it's much harder to solve tasks locally.

[aegis_camera]

Here is a reference https://www.sharpai.org/benchmark/ For specific tasks, local model could achieve workable level.