Inworld AI
Real-time voice AI stack — TTS, STT, and an LLM router
Overview
A full-stack real-time voice AI platform: sub-200ms TTS with voice cloning, streaming STT with speaker and emotion profiling, a full-duplex Realtime API, and an LLM router across 220+ models. Built for game NPCs, AI companions, and voice agents; used by NVIDIA, Disney, Ubisoft, and Xbox.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 13 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- TTS-1.5 Mini ~$5/1M chars
- Realtime TTS-2 ~$10/1M chars
- STT from ~$0.15/hour
- ~20M TTS chars at $15/1M
- Up to 3,000 custom voices
- Higher API limits
- ~120M chars at $12.50/1M
- Up to 30,000 custom voices
- Compliance add-ons
- Rates as low as $5/1M chars
- Custom limits and SLAs
What you can produce with Inworld AI
- Realtime TTS, sub-200ms, with voice cloning from 5-15s samples
- Streaming STT with speaker, emotion, pitch, and accent profiling
- Full-duplex speech-to-speech Realtime API over WebSocket
- LLM Router across 220+ models (OpenAI/Anthropic-compatible)
- Open-source SDK, Python/JS examples, and runtime templates
- Self-hostable LiveKit Agents fork
- #1-ranked TTS-2 on the Artificial Analysis Speech Arena
ASEAN Perspective
Inworld AI in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Inworld has moved on from its original AI-NPC framing into real-time voice infrastructure, and the reinvention holds up. Its TTS-2 sits at the top of the Artificial Analysis Speech Arena, latency runs sub-200ms, and a client roster that includes NVIDIA, Disney, Ubisoft, and Xbox points to real production use rather than demos. The June 2026 price cuts — more than 50% across TTS, STT, and LLM routing — make it one of the cheaper premium voice stacks, with metered TTS starting near $5 per million characters and STT around $0.15 an hour. The open-source SDK and LiveKit integration let you self-host the pipeline, which matters if latency or data residency is a concern. This is developer infrastructure, not a consumer app: expect real integration work, and expect to weigh it against ElevenLabs, Deepgram, and OpenAI in a crowded, fast-moving market.
What people say
The story since late 2025 is a pivot. Inworld built its name on drop-in character brains for game NPCs; by 2026 it markets a real-time voice AI stack — TTS, STT, a speech-to-speech Realtime API, and an OpenAI- and Anthropic-compatible LLM Router — behind a single API. The eesel AI teardown frames it plainly as an infrastructure provider now, not a character-creation toy.
Benchmarks are the loudest selling point. Inworld's Realtime TTS-2 ranks first on the Artificial Analysis Speech Arena, and the company leans on sub-200ms latency and voice cloning from 5-15 second samples. STT adds voice profiling that reads who is speaking and how — pitch, emotion, accent — which is unusual at streaming rates near $0.15 per hour.
June 2026 brought aggressive repricing. Inworld cut costs more than 50% across TTS, STT, and LLM routing, pitching it as removing "the biggest wall in consumer AI: cost." Realtime TTS-2 starts around $10 per million characters and TTS-1.5 Mini near $5 — under half the published rate of comparable premium voices per BuildMVPfast's June 2026 comparison. Subscriptions ladder up: roughly $300/mo covers about 20M characters at $15/1M, $1,500/mo about 120M at $12.50, and enterprise commitments step down toward $5/1M at the largest volumes. Named plans (Creator, Builder, Developer, Growth) scale discount tiers and custom-voice counts from 3,000 up to 30,000.
The developer story is genuinely open. The github.com/inworld-ai org hosts 27 repositories, including the TTS training and modeling code behind TTS-1 and TTS-1-Max, Python and JavaScript API examples, runtime templates, and a fully open-source LiveKit Agents fork you can run entirely on your own servers.
The caveat is category, not quality. This is plumbing for developers building companions, agents, and support voices, and it competes directly with ElevenLabs, Deepgram, and OpenAI's voice APIs — a space where prices and rankings shift monthly, so today's benchmark lead is worth re-checking before you commit.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Inworld AI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Inworld AI unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Inworld AI directly →
Spotted something out of date? Suggest an update →
Alternatives to Inworld AI
More in Agents & Automation
