Nebius retires hosted models with notice, legacy ids need their own serving endpoint

TL;DR

Do not hardcode legacy Nebius model IDs in production: check the deprecation notice page and the Token Factory model list before each deployment, since public hosted models get retired. Pin to a currently served model id and keep the model list check in your deploy script.

Steps

1. Do not hardcode legacy Nebius model IDs in production: check the deprecation notice page and the Token Factory model list before each deployment, since public hosted models get retired. Pin to a currently served model id and keep the model list check in your deploy script. If a legacy id is required, budget for a dedicated serving endpoint rather than assuming serverless availability.

When to use

You are seeing this: Do not hardcode legacy Nebius model IDs in production: check the deprecation notice page and the Token Factory model list before each deployment, since public hosted models get retired. Use this skill when you run into "Nebius retires hosted models with notice, legacy ids need their own serving endpoint".

When not to use

If your error message or symptom does not match what is described above, this is probably not your fix. Search for your exact error text instead of forcing this one to fit.

Versions

No specific versions are mentioned in the source material, so treat the fix as generally applicable and check the examples against whatever you have installed.

Why this happens

The original report does not dig into a root cause. It documents the symptom and the fix that resolved it.

Source: A Nebius-hosted recipe documents the model retirement pattern to watch. The public hosted model for the Cosmos3 recipe was retired under an August 2026 deprecation notice, and explicit legacy model IDs require a serving endpoint: the recipe notes that it does not establish their current availability. The current path is the Token Factory guide and its migration verification.

Verified resolution: Do not hardcode legacy Nebius model IDs in production: check the deprecation notice page and the Token Factory model list before each deployment, since public hosted models get retired. Pin to a currently served model id and keep the model list check in your deploy script. If a legacy id is required, budget for a dedicated serving endpoint rather than assuming serverless availability.

Source URL: https://github.com/nebius/nebius-physical-ai/blob/HEAD/docs/hackathon-cosmos3-reasoner.md