Inworld

Inworld is a developer-first realtime voice AI platform for building voice agents at consumer scale. It bundles first-party text-to-speech, speech-to-text, and a zero-markup LLM router into one API, so you can ship full-duplex voice conversations over a single WebSocket or WebRTC connection. Teams use it for game NPCs, companion apps, customer support, education, and health & wellness experiences—anywhere natural, steerable voice matters. With Realtime TTS-2 and TTS-2 Flash delivering sub-100ms time-to-first-byte, plus voice cloning from just 15 seconds of audio and cross-lingual support across 15 languages natively (200+ languages for speech), Inworld targets builders who need production-grade voice without the usual latency or cost penalties. The platform's Realtime API handles intelligent turn-taking, mid-session function calling, and dynamic context management, so you can control context length and token spend mid-conversation. The Realtime Router gives you access to 220+ LLMs from OpenAI, Anthropic, and Google at provider rates—no markup—with built-in failover and A/B testing. More recent additions include vision and listening capabilities for multimodal agents, timestamp alignment for precise integration, and an Inworld CLI for faster workflows. For game developers, the new Unreal AI Runtime offers a unified interactive AI toolkit. Pricing is a key differentiator. On the Growth plan, TTS-2 costs $12.50 per 1M characters versus $100 on ElevenLabs, and STT drops to $0.10 per hour versus $0.46 on Deepgram. A free On-Demand tier lets you prototype with up to 70 minutes of TTS included, and paid plans from $25/mo scale with credits and volume discounts. Enterprise options include on-prem deployment, custom SLAs, and HIPAA/BAA add-ons. Inworld is API-first and assumes developer expertise—this is not a no-code builder. If you're a team that wants high-quality, steerable voice with aggressive cost efficiency, it's a compelling alternative to ElevenLabs and Deepgram.