Is Claude 3.5 Available on Vertex AI? | Deprecation Status and Migration Guide

Is Claude 3.5 Still Available on Vertex AI? — Conclusion

To cut to the chase: as of June 2026, Claude 3.5 Sonnet (claude-3-5-sonnet, both the original and v2) has been removed from Vertex AI's available model list and cannot be called for new use. On the other hand, Claude 3.5 Haiku remains in the list under the model ID claude-3-5-haiku@20241022 with a "Deprecated" status — it can still be called at this time, but is not recommended.

The most important point to understand here is that Vertex AI is not a platform operated by Anthropic itself — it is a partner platform operated by Google Cloud. As explained later, deprecation schedules are managed separately from the Anthropic direct API, so assuming that a model retired on a certain date via the direct API was also retired on the same date on Vertex AI is incorrect.

If you are building a new system, there is no reason to choose Claude 3.5. The right approach is to use the current models available on Vertex AI — Claude Sonnet 4.6 or Claude Haiku 4.5 — from the start.

Source: Claude on Vertex AI — Claude API Docs

Why Do End-of-Life Dates Differ Between Vertex AI and the Anthropic Direct API?

Anthropic's official deprecation documentation includes a clear note stating that the dates on that page apply to platforms operated by Anthropic (the Claude API, Claude Platform on AWS, and Microsoft Foundry). Partner-operated platforms such as Amazon Bedrock and Vertex AI set their own deprecation schedules, so model lifecycle statuses and dates may differ.

The end-of-life dates for Claude 3.5 on the Anthropic direct API are as follows:

1. Claude 3.5 Sonnet (claude-3-5-sonnet-20240620 / claude-3-5-sonnet-20241022): Deprecation notice issued August 13, 2025; retired October 28, 2025.
2. Claude 3.5 Haiku (claude-3-5-haiku-20241022): Deprecation notice issued December 19, 2025; retired February 19, 2026.

These are dates for the direct API only. The exact end-of-life dates on Vertex AI are managed by Google Cloud and must be verified in Vertex's Claude model list. It is important not to assume that because the direct API retired a model in October, Vertex AI did the same on the same date.

Source: Model deprecations — Claude API Docs

The Claude 3.5 Generation and Its History on Vertex AI

The Claude 3.5 generation is a family of models released by Anthropic in 2024. Google Cloud began offering Sonnet 3.5 on Vertex AI in June of that year, and expanded to an improved v2 (with simultaneous computer use support) in October. Pricing and detailed specifications were those of the time and differ from current offerings, so check the Vertex AI model list for the latest information.

Source: Announcing Claude 3.5 Sonnet on Vertex AI — Google Cloud Blog / Upgraded Claude 3.5 Sonnet with computer use on Vertex AI — Google Cloud Blog

The Current Status of Claude 3.5 in Vertex AI's Model List

Looking at the list of API model IDs for Claude models currently available on Vertex AI, the situation for the 3.5 generation is clear. One notable detail is that Vertex AI model IDs use an @ separator (e.g., claude-3-5-haiku@20241022) rather than the hyphen-separated format used by the direct API (claude-3-5-sonnet-20240620).

The current status of the 3.5 generation as read from the list is as follows:

1. Claude 3.5 Sonnet (original and v2): No longer listed in the current Vertex AI model list; new use is not possible. In the same list, Claude Sonnet 3.7 (claude-3-7-sonnet@20250219) is explicitly marked as "Retired", and Sonnet 3.5 — being an earlier generation — has already been removed from the list entirely.
2. Claude 3.5 Haiku: Remains in the list as claude-3-5-haiku@20241022 with a "Deprecated" label. Deprecated means "still functional but not recommended, and will eventually be assigned an end-of-life date."

In other words, the situation for "Claude 3.5" on Vertex AI is asymmetric: Sonnet is effectively unavailable, while Haiku alone is lingering in a deprecated state. To check the latest status in your own environment, search for "Claude" in Vertex AI's Model Garden, or check the end-of-life dates in Google Cloud's Claude model documentation.

Source: Claude on Vertex AI — Claude API Docs / Anthropic Claude models on Vertex AI — Google Cloud

Migration Targets — Current Claude Models on Vertex AI

To migrate away from Claude 3.5, switch to the current models available on Vertex AI. The replacements officially recommended by Anthropic are as follows:

1. Claude 3.5 Sonnet → Claude Sonnet 4.6 (Vertex model ID: claude-sonnet-4-6). The successor to the mid-tier model, with significantly improved reasoning and coding performance.
2. Claude 3.5 Haiku → Claude Haiku 4.5 (Vertex model ID: claude-haiku-4-5@20251001). The successor for lightweight, low-cost use cases.

The benefits of migrating go beyond performance. On Vertex AI, Claude Opus 4.8 / 4.7 / 4.6 and Sonnet 4.6 support a 1 million token context window (other models such as Sonnet 4.5 support 200,000 tokens). This is a major expansion from the 200,000 tokens of the 3.5 generation, making it possible to handle large codebases and long documents in a single call.

In most cases, migration is as simple as swapping out the model ID, but since there are per-model specification differences such as temperature parameter handling, it is recommended to test before deploying to production.

Source: Model deprecations — Claude API Docs

How to Call Current Claude Models on Vertex AI

Here are the minimal steps to call a current model on Vertex AI after migrating from Claude 3.5:

1. Enable the Vertex AI API in your GCP project and enable the target Claude model in Model Garden.
2. Run gcloud auth application-default login locally to set up Google Cloud credentials.
3. Install the SDK (Python example: pip install -U google-cloud-aiplatform "anthropic[vertex]").
4. Initialize an AnthropicVertex client with project_id and region (recommended: global), then specify model="claude-sonnet-4-6" and send a message.
5. If calling via REST directly, include the model in the endpoint URL rather than the request body, and always specify anthropic_version: "vertex-2023-10-16" in the body.

A minimal Python sample is shown below.

from anthropic import AnthropicVertex

client = AnthropicVertex(project_id="MY_PROJECT_ID", region="global")

message = client.messages.create(
    model="claude-sonnet-4-6",  # 旧: claude-3-5-sonnet
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hey Claude!"}],
)
print(message)

The endpoint format is https://{LOCATION}-aiplatform.googleapis.com/v1/projects/{PROJECT_ID}/locations/{LOCATION}/publishers/anthropic/models/{MODEL_ID}:streamRawPredict. Vertex AI offers three types of endpoints — global, multi-region, and regional — with regional and multi-region being 10% more expensive than global. If you have no data residency requirements, global is recommended. For a more detailed explanation of enabling and authentication, also see How to Use the Claude API with Google | Vertex AI Steps and Pricing.

Source: Claude on Vertex AI — Claude API Docs

Deprecation Schedule Differences: Anthropic Direct API vs. AWS Bedrock

Claude model end-of-life dates are managed separately for each delivery platform. To summarize:

1. Anthropic direct API, Claude Platform on AWS, Microsoft Foundry: Operated by Anthropic. The dates on the deprecation page apply. Sonnet 3.5 retired 2025-10-28; Haiku 3.5 retired 2026-02-19.
2. Vertex AI, Amazon Bedrock: Partner-operated. Each provider sets its own deprecation schedule, so dates may diverge from the direct API.

Without understanding this structure, it is easy to make mistakes such as assuming Vertex AI follows the direct API's schedule, or conversely believing that because a model still works on Vertex, it must also be safe on the direct API. When planning a migration, always verify the end-of-life date in the model list for the platform you are actually using. For the Anthropic direct API, you can export usage by API key and model from the Console's Usage screen; for Vertex AI, you can track the models being used via Cloud Logging request and response logs.

Source: Model deprecations — Claude API Docs

Frequently Asked Questions (FAQ)

Q. What happens if I specify claude-3-5-sonnet on Vertex AI? A. Since Sonnet 3.5 has been removed from Vertex AI's model list, new calls will fail. Replace it with a current model such as claude-sonnet-4-6.

Q. Is Claude 3.5 Haiku still available? A. It remains as claude-3-5-haiku@20241022 in Deprecated status and can currently be called. However, it is deprecated and will eventually be retired. For new development, use claude-haiku-4-5@20251001 (Haiku 4.5).

Q. I was using computer use from Sonnet 3.5 v2. Is it available in the successor models? A. Tool use including computer use (screen interaction) is available on Vertex AI with current Claude models as well. The functionality has expanded compared to the 3.5 v2 era.

Q. How much code will I need to change for the migration? A. In most cases, just swapping the model ID is enough. The key things to know for Vertex AI are that IDs use the @ separator (e.g., claude-haiku-4-5@20251001), and that the anthropic_version: vertex-2023-10-16 specification remains unchanged.

Whether you are building a new system or extending the life of an existing one, there is virtually no reason to choose Claude 3.5 as of 2026. Migrating to the current models on Vertex AI as soon as possible is the most rational choice across all three dimensions: performance, context length, and support.

Source: Claude on Vertex AI — Claude API Docs