The Ember-1 model is currently being actively discussed on Hacker News. The research team further trained Kimi K3 so that the model reasoned about 40% shorter. 😳 The quality remained at the same level, and the model itself became about 40% faster and cheaper. And this is a good example of how it makes sense to build advanced LLMs today. If you have tens of millions of dollars, you don't necessarily need to train a new model from scratch. You can take an already strong existing model and invest this budget in additional training. https://fireworks.ai/blog/ember-1 Ember-1 is available now on Cline, including the new desktop app.