A speech-generation service produces expressive, low-latency audio with voice cloning and controllable delivery for interactive applications.

A speech-to-speech API connects speech recognition, language models, and voice generation for interactive real-time conversations.

LLM inference API purpose-built for consumer AI apps such as tutors, companions, and games. Serves open models including Gemma 4, GLM, DeepSeek, and MiMo with optimizations for time-to-first-token, KV-cache-aware routing, automatic fallbacks, and cheap cached input.

A unified model gateway selects and routes requests across language models based on user context, cost, quality, and availability.

A streaming transcription service converts speech into text with speaker labels, timestamps, vocabulary customization, and contextual voice signals.