AI built
for people

Inworld AI makes realtime voice models and LLM inference for consumer applications: companions, learning, health and fitness, social apps, personal agents and entertainment. The voice models are Realtime TTS-2 and Realtime TTS-2 Flash for text to speech, Realtime STT-1 for speech to text, and speech to speech, one connection that listens, reasons and talks back. Inference serves open models such as Gemma, GLM and DeepSeek on a fleet we operate, which handles tens of trillions of tokens a month in four regions: the US, Europe, India and Asia Pacific.

Inworld AI's Realtime TTS-2 Flash is a text to speech model built for realtime apps, with first audio in as little as 50 ms on Coval's public benchmark, up to 3× sooner than ElevenLabs, and it ties Eleven v3 in blind listening. Realtime TTS-2, the model you direct in plain words, starts speaking in as little as 90 ms at P90 and speaks about 500 languages and dialects. Both models stream audio back as it is generated.

Inworld AI's text to speech costs $25 per million characters on demand for Realtime TTS-2 and $15 for Realtime TTS-2 Flash, with lower rates on paid plans. On the Growth plan they are $12.50 and $7, and on Enterprise they go down to $10 and $5. At about 1,000 characters a minute, an hour of Flash speech costs about $0.42 on Growth. Realtime STT-1 speech to text is $0.10 an hour of audio on every paid plan, and open LLMs on Inworld Inference start from $0.07 per million input tokens.

Yes, Inworld AI's speech to speech API lets you choose the LLM for each session and runs the whole conversation over one connection that listens, reasons and talks back. It uses the same events and session settings as OpenAI's realtime protocol, so an existing client moves over with few changes. First audio arrives as fast as 204 ms after the user stops talking, with a median of 462 ms. You can also use the voice models on their own.

Inworld AI's Inference API serves open models on a fleet we operate, including GLM 5.3 Flash, Gemma 4 31B, Gemma 4 26B, MiMo 2.6 Pro and DeepSeek V4.1 Flash, through an OpenAI-compatible API. Open models start from $0.07 per million input tokens, for Gemma 4 26B. We optimize each model from the weights to the kernels on real traffic, so Gemma 4 31B's median first token became up to 2.8× faster, and 99.999% of GLM 5.3 Flash requests succeeded over 30 days.

You start building with Inworld AI by creating an API key and sending one request: text to speech with a voice and the Realtime TTS-2 model, a transcript from Realtime STT-1, or a chat completion to an open model through the OpenAI-compatible Inference API. Coding agents such as Claude Code and Cursor can add Inworld for you, because the Inworld skill gives them the APIs, models and code patterns to use. The playground lets you hear the voices and try the models before you write code.