Live video experiences
Connect video generation to a player and live controls, so a scene responds while it runs. Add speech and it becomes an avatar.
PromptGenerationPlayer
Build live video experiencesExplore the models
Prove it on
your own traffic.
Models tuned for latency-sensitive work. Point the client you run today at a uRun model API. Decreases in latency translate to shorter wait times for users and cheaper infrastructure costs for your business.
VoiceWhisper large-v3
99-language speech recognition with native translation into English text
Request access
VoiceKokoro 82M
Streaming text-to-speech in nine languages with cancelable utterances
Request access
VoiceKyutai STT 1B
Streaming transcription in English and French with turn detection built in
Request accessVoiceMoshi 7B
Full-duplex speech-to-speech that listens and speaks at once, with barge-in
Request access
VideoWan 1.3B Rolling
Continuous prompt-steered video that keeps rolling, steered as it plays
Request access
VideoLTX 2.5 Realtime
Chunked realtime video with live prompt controls and cuts between chunks
Request access
ChatQwen2.5-Coder-7B
A compact 7B coder for everyday code assistance, completions to edits
Request accessChatQwen3.8-27B
Long-context chat and reasoning across 262K tokens, thinking on by default
Request accessChatQwen3-Coder-480B-A35B
Large-scale coding on a 480B mixture-of-experts with 35B active parameters
Request accessVoice pipeline
Three models.
One fast conversation.
Speech recognition. A 27B language model. Speech synthesis.
Whisper large-v3Qwen3.8-27BKokoro 82M run together on one machine, with no queue or network trip between them. You connect to the pipeline the same way you connect to a single model.
Mic-to-speaker~160 ms
Use cases
What you can build
Two applications built from the models, one for live video and one for speech across languages. Each keeps the models separate from your application logic.
Speech across languages
Turn incoming speech into English text, with optional spoken output. See the models and stages you need.
SpeechTranslationText or voice
Build speech translation experiencesLabs
You build the model.
We help you serve it.
The platform is open for model labs. Work with us to bring your models to developers, or have our engineers tune and serve your own.
Bring your model to the collection
Make your model available to developers on uRun. Your weights stay under your control, and distribution is opt-in.
Discuss a model partnership
Your weights. A team to help run them.
Have your own model or pipeline? Run it on uRun and work with our engineers on the configuration your application needs.
Discuss custom model servingBuild with uRun
Find your model.
Build your next feature.
Explore the catalog and request access to the models your application needs, across speech, language, and video.