Baseten Launches joey-1, a New AI Model for Decision-Making

A New Approach to AI

Baseten has unveiled joey-1, an open-source model featuring 2 billion parameters. Unlike traditional chatbots designed for creative writing, this system is engineered to make specific decisions. It provides users with an answer alongside a corresponding probability score, making it ideal for classification and analysis tasks.

Technical Capabilities and Speed

The primary advantage of joey-1 is its optimization. According to Baseten, the model can process a typical 2,000-token request in approximately 20 milliseconds. This makes it 2.5 times faster than the elephant-1 model, which the company introduced just last week. Such performance is critical for tasks like spam identification, sorting customer support tickets, and blocking malicious requests.

Economically, joey-1 is highly efficient, costing only $0.01 per million tokens. For example, if a company needs to process 10 million support tickets per month, using joey-1 would cost approximately $50, whereas larger models could drive that expense up to $500.

Model Foundations and Competition

Built on Alibaba’s Qwen3.5-2B architecture, joey-1 was trained on 6.21 million decision examples. While other models with similar functionality exist—such as a version developed by Microsoft that costs $0.042 per million tokens—Baseten claims its product is the most powerful decision-making model in the sub-3-billion parameter category. The model weights are available on Hugging Face under an MIT license.

This development is significant for technology companies, including those in Georgia, that are integrating AI into their automation processes. Reduced costs and high processing speeds allow for more complex services to be delivered at a lower price point, facilitating the scaling of AI solutions within the local market.