Skip to content

Inference

Verda Inference is the easiest way to use open models on Verda. The API is OpenAI-compatible: if you already use an OpenAI client, point it at Verda's base URL and use your Verda API key.

Preview

Verda Inference is in preview. It's live and ready for real workloads. We're defining SLAs and SLOs for uptime and performance, usage limits apply, and details may change before general availability. Send feedback through the chat in the Console.

We don't use your prompts or responses to train models, and we don't store them. See Your data.

Get started

  • Quickstart

    Enable a model, create an API key, and make your first call.

    Quickstart

  • Models

    The models you can call, their model IDs, and how to enable them.

    Models

  • OpenAI compatibility

    Use the OpenAI SDKs with Verda: streaming, reasoning output, and errors.

    OpenAI compatibility

  • Coding agents

    Use Verda models in OpenCode and Pi Coder.

    Coding agents

Models

Model Provider Model ID
Kimi K3.1 Pro Moonshot AI kimi-k3-1-pro

You'll find the full list, with prices and live status, under Inference > Models in the Console.

Rate limits

Each project has its own requests-per-minute and tokens-per-minute limits for each model. See Rate limits.

Where it runs

Models run on Verda GPUs in our datacenters in Europe. Each model's page lists the country it runs in.

Pricing

You pay per token, from your project balance:

  • Input: the tokens you send.
  • Cached input: input tokens that repeat the start of a recent request, such as an agent resending its history. They cost less on models that offer a cached input price. Caching is automatic, and context stays cached for up to 2 hours.
  • Output: the tokens the model generates, including reasoning.

Each model's page in the Console shows its prices per 1M tokens, in your billing currency. Every model runs on the Standard tier.

Requests that fail or are interrupted, including timeouts, client disconnects and streams that end early, aren't charged.

Your project needs a balance of at least $1 to enable a model. The minimum top-up is $20. If the balance runs low, inference is paused until you top up. You can follow token usage under Inference > Usage and costs under Billing > Billing history > Inference.

Your data

  • Prompts and responses aren't stored. Request context is held briefly in an in-memory inference cache and is removed within 2 hours. We keep only the operational metadata needed to run the service and bill you.
  • No training. We don't use your inputs or outputs to train any AI model.
  • Nothing is shared with the model provider. Models run on Verda's own infrastructure, and no request data is sent to the model developer or any other third party.

See the FAQ for more, and the Trust Center for our certifications.

When to use something else

Terms

Inference is covered by the Terms of Service, including the AI Terms of Service in Schedule 5, and by each model's own license, linked from its model page. The Privacy Policy explains how we handle personal data.