Text-to-SpeechAnonymized

Inworld TTS-1.5 Max

Inworld TTS-1.5 Max is the highest-ranked text-to-speech model globally, delivering ultra-low latency, expressive speech, and zero data retention for production-grade voice AI.

Maker
Inworld AI
Modality
Speech
License
Proprietary
Open weights
No — proprietary

Overview

What is Inworld TTS-1.5 Max

Inworld TTS-1.5 Max is a state-of-the-art text-to-speech model developed by Inworld AI, released in January 2026. It delivers ultra-realistic, context-aware speech with sub-250ms P90 latency, supports over 200 languages, and leads independent benchmarks in voice quality with an Elo of 1,197–1,236.

Running it privately on Venice

On Venice, Inworld TTS-1.5 Max runs under an anonymized privacy tier—your text prompts are never stored, profiled, or used for training. This means developers get enterprise-grade voice synthesis with full data sovereignty, ideal for sensitive applications. The model’s API-first design integrates seamlessly into privacy-conscious workflows without compromising on speed or expressiveness.

AnonymizedNo prompt trainingTEE · hardware enclaveEnd-to-end encrypted

Assessment

Strengths and limitations

Strengths
  • Ranked #1 in global speech benchmarks (Artificial Analysis Speech Arena), outperforming ElevenLabs and Gemini Flash.
  • Ultra-low latency (<250ms P90) ideal for real-time voice agents and interactive media.
  • Supports over 200 languages and locales with high multilingual fidelity.
  • High-quality instant voice cloning via API with only 5–15 seconds of reference audio.
  • 30% more expressive and 40% fewer word errors than prior generations.
Limitations
  • API-only access: no consumer interface, studio, or free tier.
  • Proprietary and closed model: cannot be self-hosted or fine-tuned.
  • Higher cost per character than budget models like Kokoro TTS.

Samples

Sample outputs

Generated on Venice with our standard prompt suite — the same scripts we run through every model of this type, so you can judge it like-for-like.

Narration

On Venice, your prompts are processed privately and never stored, profiled, or used to train anyone's model.

Conversational

Wait — so I can run a private voice model with zero data retention, and pay only for what I use? That's genuinely useful.

Expressive range

Three… two… one… liftoff! The rocket roared into the night sky as the crowd erupted in cheers.

Compare every speech model on these scripts

Specifications

Datasheet

Maker
Inworld AI
Released
January 21, 2026
Modality
Text-to-speech
Supported languages
200+
Latency
<250ms P90, ~200ms median
Voice cloning
Yes — instant via API
Open weights
No — proprietary
Privacy on Venice
Anonymized — prompts not stored
Available on Venice since
Apr 2026
License
Proprietary

API

Call it from your code

Venice exposes an OpenAI-compatible API. Point your base URL at Venice and pass the model id.

curl https://api.venice.ai/api/v1/audio/speech \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tts-inworld-1-5-max",
    "input": "On Venice, your prompts are processed privately.",
    "voice": "af_sky",
    "response_format": "mp3"
  }' --output speech.mp3

Pricing

What it costs on Venice

Billed per character on Venice: $12.50 per 1M characters of synthesized speech.

Characters / 1M
$12.50
Per 1M characters

New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.

Alternatives

How it compares

ModelBest forPrice (Venice)LatencyVoice cloningOpen weights
Inworld TTS-1.5 MaxTop-ranked TTS model with best-in-class expressiveness and multilingual support.$12.50 / 1M chars<250ms P90YesNo
ElevenLabs Turbo v2.5Faster TTFB but lower quality ranking; stronger emotional control features.$62.50 / 1M chars<100msYesNo
Gemini 3.1 Flash TTSHigh latency for class; expensive but tightly integrated with Google ecosystem.$187.50 / 1M chars~200msYesNo
Kokoro Text to SpeechMost cost-efficient and open-source, but lower quality and fewer languages.$3.50 / 1M charsYesYes

Top-ranked TTS model with best-in-class expressiveness and multilingual support.

Use cases

What it is good for

  1. 01Real-time conversational AI agents requiring natural, low-latency voice responses.
  2. 02Multilingual customer service bots with expressive, human-like delivery.
  3. 03Interactive gaming NPCs with dynamic, context-aware dialogue.
  4. 04Live translation and dubbing for video content.
  5. 05Voice cloning for personalized audio experiences in apps and media.

Prompting

Getting better results

Use natural language steering to control tone, pacing, and emotion in the generated speech.

For cloning, provide clean, 10-second reference clips in the target language for best results.

Leverage streaming WebSocket endpoints for real-time agent applications to minimize perceived latency.

Version history

Inworld Realtime TTS
2024

Initial generation model

Inworld TTS-1.5 Max
2026-01

Current flagship — faster, more expressive, top-ranked

FAQ

Frequently asked questions

Inworld TTS-1.5 Max is a high-performance text-to-speech model released in January 2026 by Inworld AI. It delivers expressive, low-latency speech synthesis with support for over 200 languages and is ranked #1 in independent voice quality benchmarks.

On Venice, Inworld TTS-1.5 Max costs $12.50 per 1 million characters of synthesized speech, billed per character with no subscription required.

No. Inworld TTS-1.5 Max is a proprietary model developed by Inworld AI. It is not open source or freely available for self-hosting. Access is via API only.

Yes. The model supports instant voice cloning via API using just 5–15 seconds of reference audio. This enables personalized voices for agents, avatars, and media applications.

The model achieves P90 time-to-first-audio under 250ms and a median latency of approximately 200ms, making it ideal for real-time conversational AI and interactive applications.

Inworld TTS-1.5 Max ranks higher in voice quality benchmarks and is more cost-effective, while ElevenLabs offers slightly lower latency and more granular emotional control. For pure quality and multilingual use, Inworld leads; for emotion steering, ElevenLabs has an edge.

Yes. On Venice, your text prompts are anonymized and not stored, profiled, or used for training. This ensures full privacy and data sovereignty for sensitive deployments.

The model supports over 200 languages and locales, making it one of the most comprehensive multilingual TTS systems available, ideal for global applications.

Run Inworld TTS-1.5 Max privately

No prompt logging. No data used for training.