Qwen 3.6 27B and 3.8 27B are the darlings of local inference at the moment.
The only other thing anyone is using is Qwen 3.8 Flash Next, only by memory-rich people.
Depending on which benchmarks you believe, these models (and the Ornith 1.5 finetune of Qwen 35B-A3B) are competitive at about Opus 4.5 to 4.7 level. That matches my experience in real tasks over the last few months.
Not bad for something you can run at home for a couple of thousand dollars.
I skimmed chad the other day. There's very little to it (by design). afaics you should be able to replicate what chad does by copying the chad system prompt into a SYSTEM.md for Pi. Then you could use it with whatever model/provider you want.
pi is a fantastic harness! they are a good default in the same way llama.cpp is. it works with everything and that is the point.
i was steering chad in the opposite direction. one model & one set of silicon -> taken to the max. swap out your CHAD_MODEL and it still runs, you just leave the drafter and the kernels behind.
Talking, sure. But have they created any good pelicans to train on? Pelicanmaxxing is clearly not a thing, which can be demonstrated by asking it for something else, or migrating to another domain (voxels, 3D modeling, and other things an LLM is supposed to be bad at, plus an unlikely combination of concepts). Looking at how relatively good for an LLM Astra is at Blender, it's obvious it was either purposefully trained for better spatial performance, or just generalizes better.
I actually think the pelican and similar tests are much less useful than it seems, but not because of pelicanmaxxing. They are supposed to serve as a vibe check of model's out-of-distribution performance, but do nothing to disentangle the generalization and memorization, which is the hard part. The combination of both is still useful though, and if you look at pelicans over time you'll see their quality is more or less correlated with that, with the exception of models specifically trained to produce vector graphics and 2D layouts.
reply