Sorry I couldn’t really figure out if this was a harness, a fine tuned model, or both. Can we use Qwen with this for example? Is the performance expected to be better in that case?

The model was the default gemini-3-flash-preview.

Harness was https://www.npmjs.com/package/dirac-cli

Since Dirac is Cline's heavily modified fork, it supports all models Cline supported, including Qwen and all popular open/closed models

As a matter of fact, I am trying to run terminal bench 2.0 using some OSS models at the moment but the slow inference speeds are causing tasks to timeout