Deprecations

Together AI regularly updates the platform with new open-source models. This page describes the model lifecycle policy and lists active redirects and scheduled deprecations.

​

Model lifecycle policy

Together AI follows a structured approach to introducing new models, upgrading existing models, and deprecating older versions, so you can rely on predictable behavior.

​

Model upgrades (redirects)

An upgrade is a model release that is materially the same model lineage with targeted improvements and no fundamental changes to how developers use or reason about it. A model qualifies as an upgrade when one or more of the following are true (and none of the “new model” criteria apply):

• Same modality and task profile (e.g., instruct → instruct, reasoning → reasoning).
• Same architecture family (e.g., DeepSeek-V3 → DeepSeek-V3-0324).
• Post-training or fine-tuning improvements, bug fixes, safety tuning, or small data refresh.
• Behavior is strongly compatible (prompting patterns and evals are similar).
• Pricing change is none or small (≤10% increase).

Outcome: The current endpoint redirects to the upgraded version after a 3-day notice. The old version remains available via dedicated endpoints.

​

New models (no redirect)

A new model is a release with materially different capabilities, costs, or operating characteristics, so a silent redirect would be misleading. Any of the following triggers classification as a new model:

• Modality shift (e.g., reasoning-only ↔ instruct/hybrid, text → multimodal).
• Architecture shift (e.g., Qwen3 → Qwen3-Next, Llama 3 → Llama 4).
• Large behavior shift (prompting patterns, output style, or verbosity materially different).
• Experimental flag by provider (e.g., DeepSeek-V3-Exp).
• Large price change (>10% increase or pricing structure change).
• Benchmark deltas that meaningfully change task positioning.
• Safety policy or system prompt changes that noticeably affect outputs.

Outcome: No automatic redirect. Together AI announces the new model and deprecates the old one on a 2-week timeline (both are available during this window). You must explicitly switch model IDs.

​

Active model redirects

The following models are redirected to newer versions. Requests to the original model ID are automatically routed to the upgraded version:

​

Deprecation policy

*Depends on usage and whether a newer version of the model is available.

• If you use a model scheduled for deprecation, you receive an email notification.
• All changes appear on this page.
• Each deprecated model has a specified removal date.
• After the removal date, the model is no longer available via its serverless endpoint, but migration options are described below.

​

Migration options

When a model is deprecated on the serverless platform, you have three options:

1. On-demand dedicated endpoint (if supported):
   • Reserved solely for you. You choose the underlying hardware.
   • Charged on a price-per-minute basis.
   • Endpoints can be dynamically spun up and down.
2. Monthly reserved dedicated endpoint:
   • Reserved solely for you.
   • Charged on a month-by-month basis.
   • Can be requested via this form.
3. Migrate to a newer serverless model:
   • Switch to an updated model on the serverless platform.

​

Migration steps

1. Review the deprecation table below to find your current model.
2. Check if on-demand dedicated endpoints are supported for your model.
3. Decide on your preferred migration option.
4. If you choose a new serverless model, test your application thoroughly before migrating.
5. Update your API calls to use the new model or dedicated endpoint.

​

Deprecation history

The table below lists all deprecations, most recent first. Notes on model support:

• Models marked “Yes” in the on-demand dedicated endpoint support column can be spun up as dedicated endpoints with customizable hardware.
• Models marked “No” are not available as on-demand endpoints and require migration to a different model or a monthly reserved dedicated endpoint.

​

Recommended actions

• Regularly check this page for updates on model deprecations.
• Plan your migration well in advance of the removal date to ensure a smooth transition.
• If you have any questions or need assistance with migration, contact the Together AI support team.

For the most up-to-date information on model availability, support, and recommended alternatives, check the API documentation or contact the Together AI support team.