At Ignite 2026, Microsoft unveiled MAI Realtime 1, its first speech-to-speech model, positioned as enterprise-grade realtime voice AI that handles interruptions, quick acknowledgments and noisy multi-speaker environments. Base audio pricing is $0.01 per minute with separate token-based reasoning fees, and Microsoft claims it beats GPT Realtime 2.1 on quality, speed and price. The model is compatible with the OpenAI Realtime API, already trialed by Agoda, and slated for Dynamics 365 Contact Center and Dragon Copilot.
Speech-to-Speech
语音到语音
End-to-end voice interaction technology that converts audio input directly to audio output without intermediate text conversion.
Curated Intelligence
Cartesia launched a managed voice agent platform for enterprises, ranking #1 on the Artificial Analysis Speech Agent Arena. The system uses a cascaded architecture with co-designed voice models and agent harnesses to achieve natural speech quality comparable to native end-to-end models while delivering higher enterprise accuracy.
Startup Phonic released Phonic-1, claiming it is the only voice model capable of executing lookups and bookings while the caller is still speaking. The model scored 87.1 on the Artificial Analysis speech-to-speech index, slightly surpassing Grok Voice 3.0, with a first-audio latency of 0.61 seconds. Despite discrepancies in third-party benchmark data, Phonic secured a $27.5M Series A and powers Ashby's AI interviews.
Phonic 1 (High) debuts at #1 on the Artificial Analysis Speech-to-Speech Index with a score of 87.1, narrowly surpassing Grok Voice 3.0. The model demonstrates superior agentic performance in Tau Voice benchmarks (79.9%) while maintaining competitive latency and cost efficiency.
ElevenLabs released two new speech-to-speech models, Eleven Realtime v1 Max and Mini, designed to optimize response speed and accuracy in AI conversations. The Max variant achieves a 98.8% score on the Big Bench Audio benchmark with a 0.57-second latency, while the Mini focuses on cost-effective natural interactions and turn-taking detection. Both are accessible via the ElevenAgents platform and an OpenAI-compatible Realtime API.
SpaceXAI has released Grok Voice 3.0, claiming it as the world's most intelligent speech-to-speech model with a top score of 86.9 on the Artificial Analysis leaderboard. The model achieves sub-second latency and outperforms competitors in enterprise benchmarks like τ-Voice and EVA while maintaining lower operational costs. It is already deployed for Tesla vehicle interactions and Starlink customer support, with API pricing remaining unchanged at $0.08 per minute.
SpaceXAI's Grok Voice 3.0 debuts at #1 on the Artificial Analysis Speech-to-Speech Index with a score of 86.9, surpassing GPT-Live-1 and Gemini 3.8 Live. The model achieves the highest task success rate (94.8%) with competitive latency, priced at $0.08 per minute, significantly undercutting major competitors.
xAI has released Grok Voice 3.0, which tops the Artificial Analysis Speech-to-Speech Index with a score of 86.9, surpassing OpenAI and Google. The model demonstrates a significant leap in task success rates for simulated customer support scenarios, rising from 56.5% to 71.6%, while offering 40% faster response times and lower costs at $0.025 per task.
Hugging Face and Cerebras have partnered to integrate the Gemma 4 model into a low-latency inference platform, creating an open, modular architecture for real-time voice interaction. The solution employs a cascaded speech processing approach to significantly reduce response times, addressing the critical latency bottleneck in current voice AI user experiences.
Hugging Face introduces a fully local runtime for the Reachy Mini robot using a cascaded VAD-STT-LLM-TTS pipeline via 'speech-to-speech'. This setup exposes a Realtime API-compatible WebSocket, enabling low-latency, privacy-preserving offline conversations without cloud dependency.