Skip to main content
Deprecation means phasing out older models or endpoints in order to provide access to newer, more capable ones. When a model or endpoint is scheduled for deprecation, we will announce it along with a shutdown date, after that date, it will no longer be available. This may require updates to your applications using Nebius AI Stufio to ensure they continue to function. As always, newer and improved versions of these models are already available in Nebius AI Studio.
We recommend benchmarking a replacement in the Playground and updating your API calls once you’re satisfied with the results.

​
Deprecation Process

When a model is scheduled for deprecation, we follow a structured process: 1. Announcement
  • Email notification sent to all affected users.
  • Documentation updated on the Deprecation page with recommended replacement model(s).
2. Transition Period (7 days usually)
  • The model continues to work as normal.
  • Technical support remains available to assist with migration.
  • We recommend testing workloads with the replacement model during this time.
3. End of Life
  • After the deprecation date, the model becomes inaccessible.
  • Requests to deprecated model IDs will return errors.

​
Customer Best Practices

  • Check the Deprecation page regularly for updates.
  • Test replacement models well before the shutdown date.
  • Plan migrations in line with the published timeline.
  • Where possible, design systems to be model-agnostic to simplify future changes.

​
Deprecation List

Deprecated ModelShutdown Date
DeepSeek:
deepseek-ai/DeepSeek-V3		Nov 3 2025
Nous Research:
NousResearch/Hermes-3-Llama-405B		Nov 3 2025
Qwen:
Qwen/Qwen3-30B-A3B
Qwen/Qwen3-14B
Qwen/QwQ-32B-Lora
Qwen/QwQ-32B-Fast
Qwen/QwQ-32B
Qwen/Qwen3-32B-Lora
Qwen/Qwen2.5-72B-Instruct
Qwen/Qwen2.5-72B-Instruct-Lora		Nov 3 2025
Mistral:
MistralAI/DevStral-Small-2505		Nov 3 2025
Meta / Llama:
Meta-Llama/Llama-3.1-405B-Instruct
Meta-Llama/Llama-3.1-8B-Instruct-Fast-Lora
Meta-Llama/Llama-3.2-3B-Instruct-Lora		Nov 3 2025
If your production workloads require dedicated, long-term availability, we now offer single-tenant endpoints with:
  • Guaranteed performance and isolation
  • 99.9% SLA uptime
  • Predictable latency and autoscaling throughput
You can reach out to our team here to set up a dedicated deployment tailored to your requirements. If you need any help choosing replacements or testing your integration, simply send an email to ai-studio-support@nebius.com to connect with our solutions team.