Build with Valen Team
Open models, benchmarks, and tools for fast multimodal decision-making. Choose a project to get started.
Run a multimodal decision model. Work with text, images, and video, and get structured probabilities.
JevalKit ↗Evaluate decision models through a common interface, prepared datasets, and reproducible scoring protocols.
Find your starting point
| I want to… | Start here |
|---|---|
| Run a Valen model locally | Local quickstart |
| Serve a local HTTP API | SGLang · vLLM |
| Call a hosted Valen model | API quickstart |
| Understand the request format | Data format |
| Evaluate a decision model | Run your first evaluation |
| Add a model to JevalKit | Model integrations |
| Understand benchmark scores | Benchmark protocols |
| Find model weights and datasets | Resources |
The decision interface
Text, images, and video → task instructions and candidates → probabilities.
Valen supports three question types: Choice for named candidates, Noul for true/false decisions, and Score for ordered grades. JevalKit evaluates these outputs using each benchmark's documented protocol.
Explore the ecosystem
Contribute
Found a documentation issue, a missing integration, or a useful new task? Read the contribution guide.
The hosted API names true/false questions predicate and returns probability; local inference and evaluation retain the noul naming. Use the API reference for hosted request and response fields.