Tests by @sadiq.toao.com and @jon.recoil.org with Qwen3 show it's a monster local LLM model that outperforms on our @cst.cam.ac.uk OCaml Tripos exercises. I wouldn't say I'm panicking, but there are small beads of sweat forming on my brow as I ponder future exam strategies https://toao.com/blog/ocaml-local-code-models