Skip to main content

Model Lifecycle

Every model in the Model Graph registry has a status that reflects where it is in its lifecycle. Understanding these stages is essential for keeping your AI integrations current and avoiding breaking changes.

Status Values​

Active → Latest → Deprecated → Sunset

Active​

The model is available and functional. It was once the latest version in its family but has since been superseded by a newer release. It continues to work normally — you can still use it in production — but it's no longer the recommended choice.

What to do: Consider upgrading to the latest model in the family when convenient. There's no immediate urgency, but newer models typically offer improved performance.

Latest​

The current recommended version in its model family. When someone asks "what's the best Sonnet model?" — this is it. Each family has exactly one latest model at any time.

What to do: You're on the most current version. No action needed.

Deprecated​

The model has been announced for removal by its provider. It still works today, but it has a deprecation_date and possibly a sunset_date indicating when it will stop functioning. Some providers give months of notice; others give weeks.

What to do: Plan your migration. Use the upgrade endpoint to find the recommended replacement. The resolve endpoint can give you a format-preserving replacement string for any alias format.

Sunset​

The model has been shut down and no longer accepts requests. API calls using this model ID will fail. This is the terminal state.

What to do: If you're still referencing this model, you need to upgrade immediately. The registry retains sunset models for historical reference and to help tools detect stale references in code.

Lifecycle Timeline Example​

Here's how the Claude Sonnet family has evolved:

ModelStatusReleaseDeprecationSunset
Claude 3 Sonnetdeprecated2024-02-292025-01-152025-07-01
Claude 3.5 Sonnetactive2024-10-22——
Claude Sonnet 4.6latest2025-10-15——

Successor Chains​

When a model is deprecated, the registry tracks its successor — the model that replaces it. In most cases, the successor is simply the latest model in the same family. However, sometimes providers recommend a specific replacement that isn't the obvious next version (for example, a model in a different family).

The successor_model_id field on a model explicitly defines this chain. When it's set, the upgrade endpoint follows it instead of defaulting to latest.

Querying by Status​

You can filter models by status using the models list endpoint:

# Find all deprecated models (migration candidates)
curl "https://api.modelgraph.ai/api/v1/models?status=deprecated"

# Find the latest version of every family
curl "https://api.modelgraph.ai/api/v1/models?status=latest"

# Find sunset models (broken references in code)
curl "https://api.modelgraph.ai/api/v1/models?status=sunset"