veri

Build and deploy open-source models for your unique use-case

Create your own model, deploy your agent in minutes.

Search...⌘K
Credits: $0.00
Workspace overview

Dashboard

Welcome back, Daniel

Welcome to Veri

Get your account ready to train and deploy models.

✓Create an account○Add a payment method○Add credits○Launch your first training job

Getting started with the API

Start building with Veri in minutes.

→Get an API Key→Go to documentation→Quickstart: Test a deployment→Use the SDK

Try a workflow

Jump into common Veri entry points.

→Test your deployed model→Try reinforcement fine-tuning→Supervised fine-tuning→Deploy your custom model

Recent training jobs

live
math-reward-grpoRUNNINGQwen/Qwen3-4B · MI300X · 82%
agent-helpfulnessEVALUATINGLlama-3.1-8B · H100 · 94%
video-caption-sftCOMPLETEDWan2.1 · A100 · 100%

Train models on your data.
Purpose-built for teams that need specialized models, Veri handles all the complexity.

1.

Multi-GPU training

Submit a job and Veri provisions the exact GPU config you need.

2.
reward+0%

Observability

Integrated logging and full visibility across training and deployment.

3.

Ease of use

Get started with only a few lines of code. We also have a CLI + MCP coming out soon on our roadmap!

Build literally anything.

We currently support 103 models: 91 language models, 9 video generation models, and 3 VLA models. Take any supported base model and supercharge it on your data.

View demos

  • Built for speed.

    Upload your dataset, pick a base model, define your reward function, and hit go. Veri handles the training loop so you can iterate faster, compare results sooner, and ship a better model ahead of schedule.

    Training Pipeline

  • Watch your model learn in real-time.

    Reward curves, KL divergence, throughput, GPU utilization — all streaming live. Catch policy collapse early and know exactly when your model converges.

    Observability

  • Competitive Pricing.

    Veri automatically selects the cheapest available provider for your requested GPU type. A100, H100, or MI300X — pay per second, no commitments, no idle costs.

    Multi-Provider Compute

Our Roadmap

Take a peek at what's coming and help us shape the future of AI training infrastructure.

view the roadmap
Read the docs →

Your model, your data.

Start training today.

Get Started