This new release (BrainSoup 1.27.0) is a major update which introduces the following changes:
- GPT-4o mini! We've added support for the impressive new model GPT-4o-Mini. This model is smarter than GPT-3.5-Turbo, has a huge 128K token context window (vs 16K for GPT-3.5-Turbo), and is 3 times cheaper to run! We strongly recommend upgrading all your GPT-3.5-based agents to GPT-4o-Mini for massive savings and improved performance.
- Speech to text input! We've added a new feature that lets you speak to your agents by voice. This function is available via a button next to the input field, or via the keyboard shortcut 'Ctrl+S'. It is based on the OpenAI Whisper model, which is highly accurate and takes into account the context of the conversation.
- Mistral Nemo: We've added support for the new open-mistral-nemo model, a very interesting uncensored 12B LLM with a very large context window of 128K tokens, which is smarter than open-mixtral-7B, and only 20% more expensive to run.
- Improved overall cost-effectiveness: We now use GPT-4o-Mini as the base model for all internal orchestration prompts in BrainSoup, resulting in a more robust and cost-effective solution.
- Improved document indexing efficiency: This is a consequence of using GPT-4o-Mini for internal knowledge extraction prompts.
- More reliable WebBrowser tool: We have fixed a problem that sometimes prevented web searches from producing results.
- Better looking login screen.
- Fixed a bug that caused the application to sometimes crash when logging out.