My 8GB GPU is now running a 27B model, and it's doing the work I used to pay ChatGPT for
The model that outlived my ChatGPT subscription
ChatGPT's monthly $20 subscription now includes features such as saved setups, adjusting model reasoning, file uploading for synthesis, and access to more advanced models. Local models, however, often lack these features and are limited to smaller, less capable versions. However, an 8GB graphics card may not seem powerful enough to change this.
But a startup named PrismML has developed a model called Bonsai 27B that comes close. This model, a compressed version of Alibaba's Qwen 3.6 27B, is 14 times smaller than the full model and can run on a small GPU. PrismML compresses the model during training, rather than compressing a finished file after training. Bonsai can handle 30-40k context and can run custom setups or 'presets' which contain instructions and reference files, features that ChatGPT's Custom GPTs will no longer have access to after December 11.
Bonsai performs well on tasks that require multi-step reasoning, such as comparing phone contracts or interpreting documents, and handles images better than other local models. With Bonsai capable of handling most of ChatGPT's core functions, the need for a subscription may no longer be necessary.
Written by urgent.news from XDA Developers's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.