Breaking

Inworld Cuts Voice AI API Prices by Up to 80% for Consumer Apps

June 10, 2026 at 12:35 EDT

Media Generation
AI Agents
Sales & Business

On June 10, 2026, Inworld AI announced an immediate repricing of its APIs, including Realtime Inference, Speech-to-Text (STT) and Text-to-Speech (TTS), delivering roughly 50% savings for most developers and over 80% in some cases. The move aims to let consumer voice apps run at scale, more cheaply and for longer, and covers the hosting of open-source models such as Gemma 4, DeepSeek and MiniMax, all served through a single OpenAI-compatible endpoint.

June 10, 2026 · Inworld AI

Voice AI Prices Cut Roughly 50% to Unblock Consumer Apps

Inworld halves rates across Realtime Inference, Speech-to-Text and its single-endpoint Realtime API — arguing that model cost, not quality, is what still blocks consumer-scale AI from being profitable.

~50%

Price cut across main voice APIs

40–95%

Cost reduction already seen by existing customers

up to 87%

Cheaper at scale, end-to-end, vs rivals

1,208

TTS-2 ELO — #1 on Realtime TTS Arena

List price → price at scale

TTS cost, per 1M characters

Realtime TTS-2

$25

list

$10

at scale

TTS 1.5 Mini

$15

list

at scale

Plus Speech-to-Text at about $0.10/hour and LLM routing across 200+ models at low cost-based markup. Tiers (Builder / Growth / Enterprise) reach up to 47% off at volume.

Why it matters

The consumer-app cost squeeze

Pricing built for enterprise

$100s / mo

Token costs sized for high-paying enterprise seats.

→

But consumer apps charge

~$10 / mo

So token consumption becomes the bottleneck.

The pitch: halve the serving cost so real users can run longer and at scale — moving consumer AI toward sustainable, profitable operation. Customers cited include Wishroll, Biblechat, Talkpal and Luvu.

One endpoint · STT → LLM → TTS

STT

with voice profiling

→

LLM Router

200+ models

→

TTS

15 languages · cloning

OpenAI-compatible, WebSocket + WebRTC, with VAD, turn-taking and interruption handling. First-audio latency cited at <200–250ms P90 (TTS-1.5 Max) and <130ms (Mini).

Developer upside

Cost efficiency now seen as mattering as much as model quality.
Cheaper serving of real users eases experimentation and scaling.
Doubled credits for new users treated as a tailwind.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.