LastChatOpenAI just released two new open-weight models, GPT OSS 20B and GPT OSS 120B, which are now available in LastChat. Both use a mixture-of-experts architecture for fast, efficient responses.
OpenAI's own benchmarks are particularly surprising. The 20B model performs on par with o3-mini, while the 120B matches o4-mini - both excellent proprietary models. For models of this size, that level of performance is impressive.
I've been testing both models and they feel snappy and capable. The 120B model in particular handles reasoning tasks well, though like most models in this size range, it occasionally struggles with very specific domain knowledge compared to the largest proprietary models.
I've also rolled out a couple of handy updates alongside these new models:
- Support for .docx and .xlsx/.xls file attachments
- Better table rendering in conversations
Try GPT OSS 20B and 120B in LastChat today and see how they compare to your current go-to models. As always, your conversations stay completely private - all chat data is stored locally in your browser, never on our servers.