ONE MODEL, EVERY PLATFORM

`llama-3.3-70b-instruct` deprecation and shutdown dates on every platform

llama-3.3-70b-instruct is tracked on 5 providers and hosting routes. First to go: Azure (Microsoft Azure / Foundry) shuts down Llama-3.3-70B-Instruct on Jan 31, 2027.

Last updated Oct 23, 2026

AzureMicrosoft Azure / Foundry

llama-3.3-70b-instruct on Azure
Date	Model	Status	Replacement	Source
Jan 31, 2027	`Llama-3.3-70B-Instruct`	Shuts down100 days	None published	Notice ↗

VertexGoogle Cloud Vertex AI

llama-3.3-70b-instruct on Vertex
Date	Model	Status	Replacement	Source
Oct 21, 2026	`llama-3.3-70b-instruct-maas`	Shut down	Alternative: `Self-deploy Llama 3.3 on Model Garden`	Notice ↗

Nebius

llama-3.3-70b-instruct on Nebius
Date	Model	Status	Replacement	Source
Aug 31, 2026	`meta-llama/Llama-3.3-70B-Instruct`	Shut down	`nvidia/Nemotron-35-Lightning`	Notice ↗

Friendli

llama-3.3-70b-instruct on Friendli
Date	Model	Status	Replacement	Source
Jun 26, 2026	`meta-llama/Llama-3.3-70B-Instruct`	Retired	None published	Notice ↗

Thinking Machines

llama-3.3-70b-instruct on Thinking Machines
Date	Model	Status	Replacement	Source
Jun 12, 2026	`Llama-3.3-70B-Instruct`	Shut down	`Nemotron-3-Super-120B-A12B, Qwen3.8-27B, or Qwen3.6-35B-A3B (non-thinking mode)`	Notice ↗

Every provider sets its own dates

A deadline on one platform says nothing about another: OpenAI, Azure, Amazon Bedrock, Google Vertex AI and model routers each retire their own copies on their own schedule. Snapshots of the same model (for example dated versions) are listed separately because each has its own date. Rows marked “No deprecation notice tracked” mean ModelClock has not seen a retirement notice for that route, not that one cannot exist.