Deprecations - as of
Vertex AI model deprecations
Google Cloud lists 54 deprecated or retired models on its official pages as of Oct 17, 2026: 27 scheduled, 2 deprecated, 25 retired. The next shutdown is gemini-2.5-flash on Oct 20, 2026, replaced by gemini-3.8-flash.
- Scheduled
- 27
- Deprecated
- 2
- Retired
- 25
- As of
- Oct 17, 2026
Google Cloud deprecation policy
Google's Model versions and lifecycle page says listed retirement dates may be extended but won't be moved earlier; long-term models stay available for at least 12 months after release, and short-term models get a fixed retirement date with at least 45 days to migrate. The open model deprecations page says the endpoint keeps working during the deprecation period, and at retirement it is permanently deactivated so API requests to a retired model ID fail. Wider service deprecations follow the Google Cloud Terms of Service Discontinuation of Services section, and the Vertex AI docs now redirect to Gemini Enterprise Agent Platform docs.
Source: Google Vertex AI (Gemini Enterprise Agent Platform) on docs.cloud.google.com, read Oct 17, 2026.
Scheduled shutdowns
27 modelsModels with a shutdown date after Oct 17, 2026, soonest first.
| Shutdown | Model | Replacement | Announced | Notes | Source |
|---|---|---|---|---|---|
| in 3 days | gemini-2.5-flashGemini 2.5 Flash | gemini-3.8-flash | -None listed | Page names gemini-3.8-flash, gemini-3.5-flash-lite or gemini-3.1-flash-lite as replacements. | Source for gemini-2.5-flash on docs.cloud.google.com |
| in 3 days | gemini-2.5-flash-liteGemini 2.5 Flash-Lite | gemini-3.8-flash | -None listed | Page names gemini-3.8-flash, gemini-3.1-flash-lite or Gemma 4 as replacements. | Source for gemini-2.5-flash-lite on docs.cloud.google.com |
| in 3 days | gemini-2.5-proGemini 2.5 Pro | gemini-3.8-flash | -None listed | Page names gemini-3.8-flash or gemini-3.5-flash as replacements. | Source for gemini-2.5-pro on docs.cloud.google.com |
| in 4 days | deepseek-ocr-maasDeepSeek-OCR (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying DeepSeek-OCR on Model Garden instead of naming a replacement model ID. | Source for deepseek-ocr-maas on docs.cloud.google.com | |
| in 4 days | deepseek-r1-0528-maasDeepSeek-R1 0528 (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying DeepSeek-R1 on Model Garden instead of naming a replacement model ID. | Source for deepseek-r1-0528-maas on docs.cloud.google.com | |
| in 4 days | deepseek-v3.1-maasDeepSeek-V3.1 (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying DeepSeek-V3.1 on Model Garden instead of naming a replacement model ID. | Source for deepseek-v3.1-maas on docs.cloud.google.com | |
| in 4 days | deepseek-v3.2-maasDeepSeek-V3.2 (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying DeepSeek-V3.2 on Model Garden instead of naming a replacement model ID. | Source for deepseek-v3.2-maas on docs.cloud.google.com | |
| in 4 days | glm-4.7-maasGLM 4.7 (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying GLM 4.7 on Model Garden instead of naming a replacement model ID. | Source for glm-4.7-maas on docs.cloud.google.com | |
| in 4 days | glm-5-maasGLM 5 (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying GLM 5 on Model Garden instead of naming a replacement model ID. | Source for glm-5-maas on docs.cloud.google.com | |
| in 4 days | gpt-oss-20b-maasgpt-oss-20b (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying GPT OSS on Model Garden instead of naming a replacement model ID. | Source for gpt-oss-20b-maas on docs.cloud.google.com | |
| in 4 days | kimi-k2-thinking-maasKimi K2 Thinking (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying Kimi K2 on Model Garden instead of naming a replacement model ID. | Source for kimi-k2-thinking-maas on docs.cloud.google.com | |
| in 4 days | llama-3.3-70b-instruct-maasLlama 3.3 70B Instruct (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying Llama 3.3 on Model Garden instead of naming a replacement model ID. | Source for llama-3.3-70b-instruct-maas on docs.cloud.google.com | |
| in 4 days | minimax-m2-maasMiniMax M2 (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying MiniMax M2 on Model Garden instead of naming a replacement model ID. | Source for minimax-m2-maas on docs.cloud.google.com | |
| in 4 days | multilingual-e5-large-instruct-maasMultilingual E5 Large Instruct (MaaS) | -None listed | Open embedding model on MaaS; page suggests self-deploying E5 Text Embedding on Model Garden instead of naming a replacement model ID. | Source for multilingual-e5-large-instruct-maas on docs.cloud.google.com | |
| in 4 days | multilingual-e5-small-maasMultilingual E5 Small (MaaS) | -None listed | Open embedding model on MaaS; page suggests self-deploying E5 Text Embedding on Model Garden instead of naming a replacement model ID. | Source for multilingual-e5-small-maas on docs.cloud.google.com | |
| in 4 days | qwen3-235b-a22b-instruct-2507-maasQwen3 235B A22B Instruct 2507 (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying Qwen3 on Model Garden instead of naming a replacement model ID. | Source for qwen3-235b-a22b-instruct-2507-maas on docs.cloud.google.com | |
| in 4 days | qwen3-coder-480b-a35b-instruct-maasQwen3 Coder 480B A35B Instruct (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying Qwen3 Coder on Model Garden instead of naming a replacement model ID. | Source for qwen3-coder-480b-a35b-instruct-maas on docs.cloud.google.com | |
| in 4 days | qwen3-next-80b-a3b-instruct-maasQwen3 Next 80B A3B Instruct (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying Qwen3 Next on Model Garden instead of naming a replacement model ID. | Source for qwen3-next-80b-a3b-instruct-maas on docs.cloud.google.com | |
| in 4 days | qwen3-next-80b-a3b-thinking-maasQwen3 Next 80B A3B Thinking (MaaS) | -None listed | Open model on MaaS; page suggests self-deploying Qwen3 Next on Model Garden instead of naming a replacement model ID. | Source for qwen3-next-80b-a3b-thinking-maas on docs.cloud.google.com | |
| in 33 days | gemini-3.6-flashGemini 3.6 Flash | gemini-3.8-flash | -None listed | Short-term availability model. | Source for gemini-3.6-flash on docs.cloud.google.com |
| in 57 days | gemini-live-2.5-flash-native-audioGemini Live 2.5 Flash Native Audio | -None listed | -None listed | -None listed | Source for gemini-live-2.5-flash-native-audio on docs.cloud.google.com |
| in 103 days | gemini-3.7-flashGemini 3.7 Flash | gemini-3.8-flash | -None listed | Short-term availability model. | Source for gemini-3.7-flash on docs.cloud.google.com |
| in 149 days | gemini-2.5-flash-imageGemini 2.5 Flash Image | gemini-3.1-flash-lite-image | -None listed | -None listed | Source for gemini-2.5-flash-image on docs.cloud.google.com |
| in 166 days | multimodalembedding@001Multimodal Embedding 001 | -None listed | -None listed | -None listed | Source for multimodalembedding@001 on docs.cloud.google.com |
| in 166 days | text-embedding-004Text Embedding 004 | -None listed | -None listed | -None listed | Source for text-embedding-004 on docs.cloud.google.com |
| in 166 days | text-embedding-005Text Embedding 005 | -None listed | -None listed | -None listed | Source for text-embedding-005 on docs.cloud.google.com |
| in 166 days | text-multilingual-embedding-002Text Multilingual Embedding 002 | -None listed | -None listed | -None listed | Source for text-multilingual-embedding-002 on docs.cloud.google.com |
Deprecated, no confirmed shutdown
2 modelsDeprecated with no shutdown date, or with a listed date that has passed while Google Cloud still lists the model. See each note.
| Listed date | Model | Replacement | Notes | Source |
|---|---|---|---|---|
| veo-3.0-generate-001Veo 3 | veo-3.1-generate-001 | The retirement date has passed but the page still lists this model among latest available models rather than in the retired table, so availability is unconfirmed. | Source for veo-3.0-generate-001 on docs.cloud.google.com | |
| veo-3.0-fast-generate-001Veo 3 Fast | veo-3.1-fast-generate-001 | The retirement date has passed but the page still lists this model among latest available models rather than in the retired table, so availability is unconfirmed. | Source for veo-3.0-fast-generate-001 on docs.cloud.google.com |
Retired
25 modelsShut down on or before Oct 17, 2026, most recent first.
All 25 retired models
| Shut down | Model | Replacement | Notes | Source |
|---|---|---|---|---|
| claude-3-haikuClaude 3 Haiku on Google Cloud | -None listed | Partner (MaaS) model; available to existing customers only during deprecation. | Source for claude-3-haiku on docs.cloud.google.com | |
| claude-3-5-haikuClaude 3.5 Haiku on Google Cloud | -None listed | Partner (MaaS) model; available to existing customers only during deprecation. | Source for claude-3-5-haiku on docs.cloud.google.com | |
| gemini-2.0-flash-liteGemini 2.0 Flash-Lite | gemini-3.1-flash-lite | -None listed | Source for gemini-2.0-flash-lite on docs.cloud.google.com | |
| gemini-2.0-flashGemini 2.0 Flash | gemini-3.1-flash-lite | -None listed | Source for gemini-2.0-flash on docs.cloud.google.com | |
| claude-3-7-sonnetClaude 3.7 Sonnet on Google Cloud | -None listed | Partner (MaaS) model; available to existing customers only during deprecation. | Source for claude-3-7-sonnet on docs.cloud.google.com | |
| jamba-1.5-miniAI21 Jamba 1.5 Mini | -None listed | Partner (MaaS) model in Preview; available to existing customers only during deprecation. | Source for jamba-1.5-mini on docs.cloud.google.com | |
| jamba-1.5-largeAI21 Jamba 1.5 Large | -None listed | Partner (MaaS) model in Preview; available to existing customers only during deprecation. | Source for jamba-1.5-large on docs.cloud.google.com | |
| claude-3-5-sonnet-v2Claude 3.5 Sonnet v2 on Google Cloud | -None listed | Partner (MaaS) model; available to existing customers only during deprecation. | Source for claude-3-5-sonnet-v2 on docs.cloud.google.com | |
| claude-3-5-sonnetClaude 3.5 Sonnet on Google Cloud | -None listed | Partner (MaaS) model; available to existing customers only during deprecation. | Source for claude-3-5-sonnet on docs.cloud.google.com | |
| imagetextImagen image captioning and VQA (imagetext) | gemini-2.5-flash-image | -None listed | Source for imagetext on docs.cloud.google.com | |
| gemini-1.5-pro-002Gemini 1.5 Pro 002 | gemini-2.5-flash | -None listed | Source for gemini-1.5-pro-002 on docs.cloud.google.com | |
| gemini-1.5-flash-002Gemini 1.5 Flash 002 | gemini-2.5-flash-lite | -None listed | Source for gemini-1.5-flash-002 on docs.cloud.google.com | |
| claude-3-opusClaude 3 Opus on Google Cloud | -None listed | Partner (MaaS) model; available to existing customers only during deprecation. | Source for claude-3-opus on docs.cloud.google.com | |
| textembedding-gecko@003Text Embedding Gecko 003 | gemini-embedding-001 | The page marks this ID with an asterisk but shows no matching footnote. | Source for textembedding-gecko@003 on docs.cloud.google.com | |
| textembedding-gecko-multilingual@001Text Embedding Gecko Multilingual 001 | gemini-embedding-001 | -None listed | Source for textembedding-gecko-multilingual@001 on docs.cloud.google.com | |
| gemini-1.5-pro-001Gemini 1.5 Pro 001 | gemini-2.5-flash | -None listed | Source for gemini-1.5-pro-001 on docs.cloud.google.com | |
| gemini-1.5-flash-001Gemini 1.5 Flash 001 | gemini-2.5-flash-lite | -None listed | Source for gemini-1.5-flash-001 on docs.cloud.google.com | |
| textembedding-gecko@002Text Embedding Gecko 002 | gemini-embedding-001 | -None listed | Source for textembedding-gecko@002 on docs.cloud.google.com | |
| textembedding-gecko@001Text Embedding Gecko 001 | gemini-embedding-001 | -None listed | Source for textembedding-gecko@001 on docs.cloud.google.com | |
| text-bisonPaLM 2 Text Bison | gemini-2.5-flash-lite | -None listed | Source for text-bison on docs.cloud.google.com | |
| gemini-1.0-pro-vision-001Gemini 1.0 Pro Vision 001 | gemini-2.5-flash | -None listed | Source for gemini-1.0-pro-vision-001 on docs.cloud.google.com | |
| gemini-1.0-pro-002Gemini 1.0 Pro 002 | gemini-2.5-flash | -None listed | Source for gemini-1.0-pro-002 on docs.cloud.google.com | |
| gemini-1.0-pro-001Gemini 1.0 Pro 001 | gemini-2.5-flash | -None listed | Source for gemini-1.0-pro-001 on docs.cloud.google.com | |
| code-geckoCodey Code Gecko | gemini-2.5-flash-lite | -None listed | Source for code-gecko on docs.cloud.google.com | |
| chat-bisonPaLM 2 Chat Bison | gemini-2.5-flash-lite | -None listed | Source for chat-bison on docs.cloud.google.com |
FAQ
Which Vertex AI models shut down next?
As of Oct 17, 2026, 27 Vertex AI models are scheduled to shut down. The earliest dates: Oct 20, 2026: gemini-2.5-flash (replacement gemini-3.8-flash), gemini-2.5-flash-lite (replacement gemini-3.8-flash) and gemini-2.5-pro (replacement gemini-3.8-flash); Oct 21, 2026: deepseek-ocr-maas, deepseek-r1-0528-maas, deepseek-v3.1-maas, deepseek-v3.2-maas and 12 more.
How much notice does Google Cloud give before shutting down a model?
Google's Model versions and lifecycle page says listed retirement dates may be extended but won't be moved earlier; long-term models stay available for at least 12 months after release, and short-term models get a fixed retirement date with at least 45 days to migrate. The open model deprecations page says the endpoint keeps working during the deprecation period, and at retirement it is permanently deactivated so API requests to a retired model ID fail. Wider service deprecations follow the Google Cloud Terms of Service Discontinuation of Services section, and the Vertex AI docs now redirect to Gemini Enterprise Agent Platform docs.
How many Vertex AI models have been retired?
This page lists 25 retired Vertex AI models as of Oct 17, 2026. The most recent is claude-3-haiku, shut down Aug 23, 2026.
Where do these Vertex AI dates come from?
Every row is copied from Google Cloud's official pages at docs.cloud.google.com, read on Oct 17, 2026, and links to the page it came from. Policy: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/model-versions
Newsletter
Know before a model shuts down
One email per week. New model releases, deprecations and pricing changes, and what they mean for what you build.