Model Catalog

Every model the Inference API serves, with pricing, context windows, and capability flags.

Models in the catalog are Production or Preview. Preview models show a Preview badge; Production models have no lifecycle badge. Early Access is granted separately and is not listed in the public catalog. See Model Lifecycle for details.

Need a specific model? Get in touch. We're continuously expanding our catalog and are open to adding new models. For full customizability and dedicated resources, check out our dedicated inference service.

Not sure which model to pick?

Tell us what you want to build and we'll point you at the right models.

All models

Programmatic access

The catalog is also available over the API via GET /v1/model/info. It returns a data list of {model_name, model_info} entries with metadata for tooling and agents: token pricing (decimal EUR per 1M tokens), the deployed model version (the HuggingFace repo sha), the reasoning_effort values each model meaningfully accepts, and deprecation and supersession info.

curl https://api.inference.nebul.io/v1/model/info

{
  "data": [
    {
      "model_name": "zai-org/GLM-5.3",
      "model_info": {
        "key": "zai-org/GLM-5.3",
        "mode": "chat",
        "model_type": "llm",
        "max_input_tokens": 1048576,
        "input_cost_per_1m_tokens": 1.47,
        "output_cost_per_1m_tokens": 4.62,
        "cache_read_input_cost_per_1m_tokens": 0.35,
        "supports_reasoning": true,
        "reasoning_efforts": ["low", "high", "max"]
      }
    }
  ]
}

The entry above is a real response, trimmed to the interesting fields. Entries also carry model_version (the deployed model version, a HuggingFace repo sha), deprecation_date, and superseded_by_model_name when set: model_version pins the deployed revision, and the two deprecation fields mark a model scheduled for retirement. Migrate to the superseding model before the date; see Model Lifecycle for the deprecation policy and the current schedule. Fields that are unset are absent from the response.

Without an API key it returns all catalog-visible models; with an API key, the results are scoped to the key's project. The endpoint is CORS-enabled for browser use.