We're thrilled to introduce DS1, the world's fastest and most cost-effective embedding model for low-latency applications. The English version is available now on the AWS Marketplace, with multimodal and code variants coming soon.
Why DS1 is different: DS1 is a static model that runs without GPU acceleration — delivering exceptional speed while dramatically reducing costs.
The DS1 Story
Our research team initially built DS1 for our own applications, including tldr.takara.ai . The performance was so remarkable that we decided to share it with the world. Learn more about DS1's performance.
Key Use Cases
Low-Latency Applications
When milliseconds matter—in real-time speech-to-speech, speech generation, or live gameplay—DS1 delivers. Not only is latency dramatically reduced, but responses remain consistently fast with minimal variability. DS1 scales gracefully without degrading performance.
Large-Scale Data Embedding
Processing vast datasets can be costly and time-consuming. DS1 optimizes both speed and storage with vectors sized at just 512 dimensions. This means faster embedding, lower storage overhead, and significantly reduced operational costs—whether you're embedding for the first time or continuously.
What's Next?
The Takara team is committed to democratizing AI. Beyond releasing multimodal and code versions of DS1, we're working to bring this same level of performance to multimodal embeddings. Stay tuned.
