Inworld AI

Inworld AI (legal entity Theai, Inc.) is a research lab and inference provider for realtime AI aimed at consumer-facing applications. Its platform combines first-party speech models with model routing and inference capacity behind a single API and billing relationship. The product set comprises Realtime TTS (the TTS-2 research preview and TTS-2 Flash, plus TTS 1.5 Max and 1.5 Mini), Realtime STT, an end-to-end speech-to-speech Realtime API, Realtime Inference (Inworld-optimized open-source models such as Gemma 4, DeepSeek V3.2/V4 and GLM-5.1/5.2), a Realtime Router that reaches 220+ LLMs across first-party and third-party tracks (OpenAI, Anthropic, Google, xAI, Meta, Mistral, DeepSeek, Qwen, Groq and DeepInfra), and Compute, which provides dedicated capacity for traffic-heavy customers.

The Realtime API streams speech in and speech out over a single WebSocket or WebRTC connection, with full-duplex audio, context-aware turn detection with adjustable eagerness, mid-session function calling, dynamic context management, and compatibility with the OpenAI Realtime API. Speech features include instant and professional voice cloning from short reference audio, text-based voice design, natural-language voice steering across dimensions such as emotion, articulation, intonation, volume, pitch, range, speed and vocal style, non-verbal tags, word/character/phoneme alignment for lipsync, and cross-lingual voice identity across 200+ languages. Realtime STT adds diarization, custom vocabularies, voice profiling, semantic and acoustic VAD, word-level timestamps and multilingual support across 30+ languages. Deployment options span hosted cloud, self-managed VPC and on-premise, with zero-data-retention configurations for regulated workloads.

The company began with a focus on AI-driven virtual characters for games, metaverse, VR/AR and brand experiences, offering a no-code character studio and integrations with Unreal and Unity, and a Character Engine orchestrating multiple models for personality, emotion, long-term memory and dialogue animation. It later positioned around the Inworld Runtime for scaling consumer AI applications before framing itself as a realtime voice AI infrastructure provider.

Founding story

The company was founded in July 2021 by Ilya Gelfenbeyn, Michael Ermolenko and Kylan Gibbs, with a third-party profile also listing Yuvakiran Arthala as a founder. The founders previously built the conversational AI platform API.AI, acquired by Google and renamed Dialogflow, and worked on generative models at Google and DeepMind, where the team led product for LLMs. According to CEO Kylan Gibbs, the company was started because AI was accruing to business automation while consumer experiences lagged, beginning with AI agents for gaming and media.

Business model

Inworld sells API access to its speech models, Realtime API, LLM Router and dedicated compute to developers, priced per usage tier (for example a Growth plan and enterprise-scale rates) with a single billing relationship across the stack. A third-party profile describes a revenue model based on subscription services and partnerships with developers integrating AI capabilities, with scalable subscription plans; dedicated GPUs are offered on a per-GPU-hour basis.

Usage- and plan-based pricing across products: per-million-characters for realtime TTS, per-hour for realtime STT, no markup added on routed LLM spend, a stated percentage of the public rate for realtime inference, and per-GPU-hour pricing for dedicated GPUs starting from $5. Rates are described as falling further at enterprise scale.

Traction

Customers cited by the company include Status by Wishroll (reported to reach 1M users in 19 days and a 95% AI cost reduction), OtherHalf, Bible Chat, Particle, Luvu, Talkpal, Playroom, AstroBeam, Isekai Zero and Latitude. A third-party profile reports over 500,000 daily active users and clients primarily in North America and Europe across gaming, media and training. Partners and integrations referenced include LiveKit, Vapi, Stream and k-ID.

Latest developments

A company resource page published 2026-02-19 describes six products — Realtime TTS, Realtime STT, the Realtime API, Realtime Inference, the Realtime Router and Compute — including the Realtime TTS-2 research preview and TTS-2 Flash, routing to 220+ LLMs, and cumulative funding of more than $125 million. The website cites provider rates as of June 2026 in publishing price comparisons and links to an explanation of its price cuts.