Inworld AI Review

Realtime voice and inference APIs for consumer apps

★★★★★4.7/5
Top Choice 2026
Updated: October 2026
AI-Powered

Ready to Get Started?

Explore Inworld's realtime voice models and inference API, starting with pay-as-you-go pricing.

Overview

Inworld AI was known for AI game characters, but its current site describes a different focus: models and inference for consumer applications, with an emphasis on realtime voice. Customers listed on the site include Wishroll, Talkpal, Latitude and Bible Chat.

The main products are Realtime TTS-2 and TTS-2 Flash (text-to-speech), Realtime STT-1 (speech-to-text), a speech-to-speech endpoint for voice agents, an OpenAI-compatible inference API for open LLMs, and dedicated compute deployments.

Pricing is usage-based. A $0 On-Demand plan lets you pay per character of speech, and monthly plans from $25 lower the per-character rate. Game studios can still use it to voice characters, but it is now a general voice and inference API rather than a character design tool.

Key Features

Realtime TTS-2 and TTS-2 Flash

Text-to-speech in about 500 languages and dialects, with inline voice direction tags and streaming over HTTP or WebSocket. Time to first audio is listed at 100 ms (TTS-2) and 20 ms (Flash).

Voice cloning and voice design

Instant voice cloning from 3 seconds of audio, or a new voice designed from a text description.

Speech-to-speech

One WebSocket or WebRTC connection that listens, reasons and talks, following the OpenAI Realtime protocol, with turn detection, interruptions and tool calling.

Realtime STT-1

Speech-to-text in 30 languages that also returns a voice profile (age, emotion, pitch, accent).

Inference API

Open models such as Gemma, GLM and DeepSeek served through an OpenAI-compatible API.

Dedicated compute

Reserved-capacity deployments in the US, Europe, India or Asia Pacific.

Pros & Cons

Advantages

  • $0 pay-as-you-go plan with a commercial license
  • Low-latency streaming text-to-speech in many languages
  • Speech-to-speech endpoint compatible with the OpenAI Realtime protocol
  • Published per-character prices that drop on higher plans
  • Pricing page states GDPR and SOC 2 Type II on every plan

Disadvantages

  • No longer centred on authoring game characters, which earlier users may expect
  • Requires development work: it is an API, not a no-code game tool
  • Free On-Demand plan has community support only
  • HIPAA, BAA and data residency are add-ons or Enterprise-only

Pricing

Inworld publishes monthly plans with per-character text-to-speech rates (prices per 1 million characters):

On-Demand: $0

Pay as you go: $25 (TTS-2) or $15 (Flash) per 1M characters. Includes up to 70 minutes of TTS, 100 custom voices, 2 professional voice clones and a commercial license.

Creator: $25/mo

$20 (TTS-2) or $10 (Flash) per 1M characters.

Builder: $100/mo

$17.50 (TTS-2) or $9 (Flash) per 1M characters.

Developer: $300/mo

$15 (TTS-2) or $8 (Flash) per 1M characters.

Growth: $1,500/mo

$12.50 (TTS-2) or $7 (Flash) per 1M characters, with ZDR, HIPAA and BAA available as add-ons.

Enterprise: custom

Rates as low as $5 per 1M characters, SLA and DPA, EU or India data residency.

Yearly billing is offered with "2 months free". LLM inference and compute are priced separately on the models page.

Best Use Cases

Inworld AI Works Well For:

  • Voicing characters or narrators in games and interactive apps
  • Realtime voice agents and companions
  • Apps that need speech-to-text with emotion and accent signals
  • Teams that want open LLMs behind an OpenAI-compatible API

May Not Be Ideal For:

  • Designers looking for a no-code tool to write character personalities
  • Projects that need HIPAA coverage on a small budget
  • Teams without developers to integrate an API

Comparison

Inworld AI vs Alternatives

Inworld now competes mostly with other voice and inference APIs.

  • Charisma.ai: Charisma gives writers a graph story editor and Unity and Unreal SDKs. Inworld provides the voice and model APIs, and you build the story logic yourself.
  • Replika: Replika is a finished consumer companion app. Inworld lists companion apps among its customers, as the infrastructure behind them.

Screenshots & Interface

Explore Inworld Ai's interface:

Frequently Asked Questions

What does Inworld AI offer today?

Realtime text-to-speech, speech-to-text, speech-to-speech for voice agents, an OpenAI-compatible LLM inference API and dedicated compute.

Is Inworld AI still for game characters?

It can still voice game characters, but the site now presents it as voice and inference infrastructure for consumer apps rather than a character engine.

How much does Inworld AI cost?

On-Demand is $0 with pay-as-you-go rates ($25 per 1M characters for TTS-2). Paid plans run from $25 to $1,500 per month, and Enterprise is custom.

Is there a free option?

Yes. The On-Demand plan costs $0 and includes up to 70 minutes of TTS, 100 custom voices and 2 professional voice clones.

Is Inworld AI compliant with SOC 2?

The pricing page states GDPR and SOC 2 Type II on every plan, with HIPAA and BAA available as add-ons on Growth. See the vendor's security page for details.

Final Verdict

4.7/5
Excellent Choice

Inworld AI is now a realtime voice and inference provider with transparent, usage-based pricing and a $0 entry plan. It is a solid choice for developers who need fast speech in games or apps, but it is no longer a dedicated tool for designing game characters.