In an innovative move, ElevenLabs has integrated the DeepSeek R1 model with its Conversational AI platform, providing a voice to the open-source model. This collaboration aims to enhance user interaction with AI, according to ElevenLabs.

DeepSeek R1 and Its Capabilities

The DeepSeek-R1-Distill-Qwen-32B model, a distilled version of DeepSeek-R1 based on Qwen2.5, has been employed for this integration. This model is noted for outperforming OpenAI-o1-mini across various benchmarks, setting new standards for dense models. The choice of the distilled version is strategic, as the pure versions of DeepSeek R1 are not yet reliable for function calling, which is crucial for seamless AI interactions.

Utilizing Custom LLM Options

The integration leverages ElevenLabs’ Custom LLM options, which are compatible with any OpenAI-compatible API provider that supports tool use or function calling. The current implementation uses Cloudflare, although guides for other platforms like GroqCloud and Together AI are available.

Getting Started with ElevenLabs Conversational AI

To begin using this integration, users can create a new AI agent on the ElevenLabs Conversational AI app page. The Math Tutor template is recommended for showcasing DeepSeek’s reasoning capabilities. This template includes a system prompt that converts mathematical expressions into natural language, enhancing the interaction experience.

For those interested in experiencing the capabilities of DeepSeek, a demo is available on ElevenLabs’ platform, allowing users to interact with the AI agent and explore its functionalities.

Future Developments

While the current version of DeepSeek R1 does not fully support function calling, ongoing developments aim to address these limitations. Users can follow the progress of these enhancements through updates provided by the developers.

Image source: Shutterstock