Awesome, thanks for benchmarking it! It makes sense, the model is reasoning-only so thinking mode is mandatory here.

→ View original post on X — @maximelabonne