Migration
Due to new and powerful hardware, BeeChat is relocated to a new server: https://inference.mlmp.ti.bfh.ch .
2025
Model update: BeeChat v3 uses gpt-oss!
We replaced llama3.3 with the new and powerfull gpt-oss model, improving performance and generation quality.
New existing features: Model feedback
From now on, users can flag each generation with a thumbs up or thumbs down and provide additional information. These feedbacks will be used to further improve the model.
Image generation deactivated
Due to the growing demand for testing BeeChat, we decided to temporarily remove image generation capabilities of BeeChat to free up resources for text generation. In the meantime, please resort to other tools.
New exciting features: Display of energy consumption and ß-replacement added
From now on, BeeChat calculates an approximate energy consumption and price individually for each prompt and aggregated for a complete conversation, fostering the awareness of AI energy usage. Moreover, we programmatically replace all instances of "ß" in model outputs, aligning BeeChat with its deployment in the heart of Switzerland.
BeeChat counts over 50 users!
We are happy to share that we've recently passed the mark of 50 BeeChat users! We are working hard to keep text generation as smooth as possible and looking forward to new ideas for experiments!
BeeChat v2 released: Bigger model - better performance!
We recently published an improved version of the initial BeeChat model, resulting in even better performance! In detail we switched from the default llama 3.3 model variant (Q4_K_M) to a variant with more precision (Q6_K).
Early adopters start experimenting!
More and more people are starting to test BeeChat for their day-to-day work at BFH: While lecturers experiment with providing curated material to the students, administrative staff test BeeChat's capatbility to analyze and scan legal documents. Researchers use BeeChat to support data analysis. Feel free to reach out to us for your own usecase!