List of the Top AI Models for LiveKit in 2026

Mercury 2 signifies a revolutionary leap in reasoning models, particularly tailored for instantaneous voice interactions, as it can promptly respond to incoming calls. In contrast to conventional autoregressive models that often leave callers waiting in silence while they generate responses sequentially, Mercury 2 uses a diffusion large language model architecture that can produce more than 1000 tokens per second on standard NVIDIA GPUs. This extraordinary processing speed enables it to finalize a complete reasoning cycle and start speaking in a timeframe that harmonizes with the natural flow of conversation, effectively reducing the usual wait time from several seconds to around 300 milliseconds. The functionality of Mercury models revolves around converting clear text into noise, after which a traditional Transformer is trained to reverse this process and predict the original text simultaneously across all positions. By adopting a denoising strategy that processes multiple tokens concurrently, the generation process becomes more efficient, achieving speeds comparable to customized silicon on NVIDIA H100s while enhancing responsiveness in voice applications. Consequently, Mercury 2 not only improves user interactions but also establishes a new benchmark for the field of interactive voice technology, paving the way for future advancements. With its innovative design, it promises to revolutionize the way users engage with voice systems.

The Gemini Live API is a sophisticated preview feature tailored for enabling low-latency, bidirectional communication through voice and video within the Gemini system. This cutting-edge tool allows users to participate in dialogues that resemble natural human interactions, while also permitting interruptions of the model's replies through voice commands. Besides managing text inputs, the model can also process audio and video, producing both text and audio outputs. Recent updates have introduced two new voice options and support for an additional 30 languages, alongside the flexibility to choose the output language as necessary. Additionally, users are empowered to modify image resolution settings (66/256 tokens), select their preferred turn coverage (whether to transmit all inputs continuously or solely during user speech), and personalize their interruption settings. Other noteworthy features include voice activity detection, new client events for indicating the conclusion of a turn, token count monitoring, and a client event for signaling the stream's end. The system is also equipped to handle text streaming and offers configurable session resumption that retains session data on the server for up to 24 hours, while also allowing for longer sessions through a sliding context window to maintain better conversational flow. Overall, the Gemini Live API significantly enhances the quality of interactions, making it not only more versatile but also more user-friendly, which ultimately enriches the user experience even further.

Gemini 3.8 Live is Google DeepMind’s real-time multimodal voice model for developers building conversational agents that can listen, speak, reason, use tools, and complete tasks during live interactions. Unlike traditional voice systems that combine separate speech recognition, language, and text-to-speech components, Gemini 3.8 Live provides a native speech-to-speech architecture intended to create more fluid conversational experiences. The model can continue streaming an audio response while asynchronous function calls execute in the background, reducing pauses when an agent needs information from external services. This makes it suitable for applications such as customer service, scheduling, transactions, enterprise assistants, interactive training, and other workflows that require both conversation and action. Gemini 3.8 Live can consume visual context alongside audio, allowing developers to create agents that respond to what a user is saying as well as what a camera or visual input is showing. Incremental content updates let applications combine real-time dialogue with structured information and update responses as new data arrives. The model supports more than 97 languages and is designed to provide consistent accents across multilingual conversations. It also improves recognition of alphanumeric information such as confirmation codes, policy numbers, claim identifiers, technical values, and other data where transcription precision is important. For more demanding workflows, Gemini 3.8 Live Extended Thinking provides configurable reasoning that can work through complex multi-step problems in the background while the main conversation continues. Developers can access Gemini 3.8 Live through the Gemini Live API and Google AI Studio, with ecosystem integrations available through platforms including LiveKit, Pipecat, Agora, Fishjam, LangChain, Vercel, and Vision Agents.