Realtime TTS-2 is live. Built for realtime conversation that feels human. Read the Realtime TTS-2 announcement

Overview

Models

Compare Inworld models for TTS, STT, and Inference by capabilities, performance, price, and deployment options.

Choose from Inworld's models based on the capabilities, performance, price, and deployment options your application needs. You can select and customize models for your use case.

Overview of Model Offerings

Use this overview to choose a model type for your application.

  • TTS: Text-to-Speech models generate high-quality audio, such as a character's voice.
  • LLM: Large Language Models generate text from inputs, usually text; some models also accept other input types. Use them to choose in-game actions, hold conversations, generate narratives, and more.
  • Embeddings: Embeddings models convert text into high-dimensional vectors for intent detection, text similarity comparison, and retrieval-augmented generation (RAG).

TTS

Use Inworld TTS models through the Agent Runtime or API. Choose a model based on your use case, quality needs, and performance requirements.

Realtime TTS-2

Our flagship, top-ranked model — the best choice for production

  • Best quality and steerability, with natural language steering for speech that fits its context
  • Support for 200+ languages and locales
  • Ultra-low latency (100 ms TTFB*) at high concurrency
  • High quality instant voice cloning
  • Enhanced timestamps with phonetic details and visemes

Realtime TTS-2 Flash

Our fastest, most cost-efficient model — built for latency-critical, high-volume workloads

  • Our lowest latency — 20 ms TTFB*, 5× faster than inworld-tts-2
  • Lowest cost per character
  • Same 200+ languages and locales as inworld-tts-2
  • High quality instant voice cloning

* P90 time to first audio byte, measured server-side — excludes network latency.

Models overview

NameModel IDDescriptionSupported languages
Realtime TTS-2inworld-tts-2              Our most powerful model with natural language steering and the strongest multilingual capabilities — the best choice for production quality200+ languages and locales — see Languages
Realtime TTS-2 Flashinworld-tts-2-flash              Our fastest, most cost-efficient model — significantly lower latency than inworld-tts-2, ideal for real-time agents and high-volume workloads200+ languages and locales — see Languages

Using an earlier model? Previous-generation models (inworld-tts-1, inworld-tts-1-max, inworld-tts-1.5-max, inworld-tts-1.5-mini) are deprecated and not recommended for new projects. inworld-tts-1 and inworld-tts-1-max were discontinued on June 15, 2026. Requests to those two models are automatically routed to newer models. If you still use any of these earlier models, we recommend the TTS-2 family for better quality, latency, and cost.

LLM

Chat Completion

Inworld provides access to 100+ LLMs from various providers through a unified Chat Completions API.

  • Available models: Browse inworld.ai/models and select Text. For programmatic discovery, use the authenticated List models API.
  • Pricing: Visit inworld.ai/pricing for model pricing details
  • Specific model: Call a model directly using the provider/model format (e.g., openai/gpt-5.4)
  • Auto-select: Set model to "auto" to automatically pick the best model based on price, latency, or performance
  • Router: Create a router for conditional routing, A/B testing, and reusable configurations, then reference it via the model field with the inworld/ prefix (e.g., inworld/my-router)

Image models

Browse inworld.ai/models and select Image for image-generation model IDs and pricing. The Image generation guide shows how to generate and save an image, use a reference image, and choose output settings. See Image and audio inputs to ask a model to analyze an image.

Decision models

Browse inworld.ai/models and select Decisions for model IDs and pricing. The Decisions guide explains typed questions, application logic and a complete request and response.

Embeddings

Service ProviderModel IDDescription
InworldSERVICE_PROVIDER_INWORLDBAAI/bge-large-en-v1.5Great for English text
InworldSERVICE_PROVIDER_INWORLDsentence-transformers/paraphrase-multilingual-mpnet-base-v2Great for multi-lingual text
OpenAISERVICE_PROVIDER_OPENAItext-embedding-3-smallGeneral purpose, efficient embedder with support for English and multi-lingual text.

Video

The Video API generates short MP4 clips with sound from a text prompt. You can start from a first and last frame (Image-to-video) or guide the video with reference images, videos and audio (Reference-to-video). Jobs run asynchronously and are billed per second of finished video.

Browse inworld.ai/models and select Video for model IDs and pricing. See Models and parameters for the documented models' aspect ratios, prompt limits and defaults, and Image-to-video for image-input support.

Terms of Service

You may not violate the terms of service or policies of third-party model providers using Inworld's platform or your account will be subject to deactivation.