AWS Bedrock vs Azure AI Foundry vs Google Vertex AI vs Alibaba Cloud Model Studio: Which LLM Platform Should You Build On?

Picking an LLM platform used to be a side effect of picking a cloud. If your company ran on AWS, you used Bedrock. If it ran on Azure, you used OpenAI through Azure. That shortcut no longer holds. In 2026, each platform has its own signature models, its own prices for the same model, and its own agent tooling, and the differences are large enough to change what you can build.

This guide compares the four platforms most teams shortlist: Amazon Bedrock, Azure AI Foundry, Google Vertex AI and Alibaba Cloud Model Studio. It covers which models each one gives you, what the same model costs on each, where your data can stay, and how to avoid betting your product on a single one. Eden AI connects to all four platforms, so the prices and routing tests in this guide come straight from our own catalogue.

The short answer

• You need OpenAI's GPT models first: Azure AI Foundry, now called Microsoft Foundry, is their historical home.
• You need Gemini: Google Vertex AI is the only one of the four that serves it.
• You want the widest choice of third-party models, with Claude: Amazon Bedrock.
• You want Qwen and Chinese open models at the lowest list prices: Alibaba Cloud Model Studio.
• You need models from more than one of them, or want to keep the option to move: use a multi-cloud API such as Eden AI, which sits in front of all four with one key.

At a glance

Amazon Bedrock

Bedrock is AWS's managed service for foundation models, and in 2026 it has the broadest third-party catalogue of the four. On Eden AI alone, Bedrock serves 156 endpoints: Anthropic's Claude family, OpenAI models including GPT-6.1 Sol and gpt-oss, Mistral, Qwen, Kimi, DeepSeek, Llama and Amazon's own Nova models.

The other half of Bedrock is the agent stack. Bedrock AgentCore provides a runtime, memory, tools and a gateway for production agents, and in October 2026 AWS added support for the Bedrock InvokeModel APIs on the AgentCore Gateway inference endpoint, according to AWS.

Choose Bedrock if

• Your stack already runs on AWS and you want IAM, VPC and billing in one place.
• You want Claude and many open models side by side, without Gemini.
• You plan to build agents on AWS-native infrastructure.

Watch out for

• Model availability varies by AWS Region, and some models are only reachable through cross-Region inference profiles.
• Gemini is not on Bedrock.

Azure AI Foundry (Microsoft Foundry)

Azure AI Foundry, which Microsoft now calls Microsoft Foundry, is the platform most teams use to run OpenAI's GPT models inside an enterprise cloud. The catalogue on Eden AI reflects that: 69 of Azure's 78 endpoints are GPT models, with DeepSeek and Mistral alongside.

Foundry's strength is the layer around the models. Foundry Agent Service runs hosted agents, and in September 2026 Microsoft added voice agents in public preview, according to the Microsoft Foundry blog. Teams already on Microsoft 365, Entra ID and Azure get identity, governance and billing in one place.

Choose Azure AI Foundry if

• GPT models are your default and you want them under Azure's enterprise agreements.
• Your organisation runs on Microsoft identity and security tooling.
• You need EU-pinned GPT deployments.

Watch out for

• Models are deployed per model and per region, which adds setup and quota management.
• GPT is no longer exclusive to Azure: it is also on Bedrock, and in private preview on Google Cloud.

Google Vertex AI

Vertex AI is Google Cloud's AI platform, which Google now presents as its Gemini platform. It is the only one of the four that serves Gemini, and it also carries Anthropic's Claude: on Eden AI, Vertex serves 41 Claude endpoints and 39 Gemini endpoints.

Google is pushing breadth hard. On 21 October 2026 it started rolling out OpenAI's GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna in private preview in Model Garden, next to more than 200 first-party, third-party and open models, and it now describes Google Cloud as the only hyperscaler offering Gemini, Claude and GPT, according to Google Cloud.

Choose Vertex AI if

• Gemini is central to your product, for long context, multimodal input or price.
• You want Gemini and Claude on the same platform and the same invoice.
• Your data already lives in BigQuery and Google Cloud.

Watch out for

• OpenAI models on Google Cloud are in private preview for select customers, not generally available yet.
• Google now uses the Gemini platform name, while most APIs and docs still say Vertex AI.

Alibaba Cloud Model Studio

Model Studio is Alibaba Cloud's model platform. It serves the full Qwen series and a set of third-party models, including DeepSeek, Kimi and GLM, through Qwen's own APIs and OpenAI-compatible endpoints, according to Alibaba Cloud. On Eden AI, Model Studio serves 68 endpoints, 56 of them Qwen.

Its main argument is price. DeepSeek V4.1 Flash lists at $0.15 per million input tokens on Model Studio, half the Vertex AI price. Model Studio runs in Singapore, Frankfurt, Virginia, Tokyo, Hong Kong and Beijing, according to Alibaba Cloud's regions page.

Choose Model Studio if

• Qwen models are your default, from small fast models to Qwen3.8-Max.
• You run high volumes on open models and list price drives the decision.
• You serve users in Asia.

Watch out for

• Prices and available features depend on the region and the deployment scope you choose.
• Data-processing terms differ by region, so check the one you pick against your compliance needs.

Same model, different bill

The most useful comparison is not the platforms' catalogues but what the same model costs on each. Here are list prices from the Eden AI catalogue on 29 October 2026, per million tokens.

Three things stand out. No platform is cheapest across the board: Vertex AI wins on gpt-oss-120b and loses on DeepSeek V4.1 Flash. Frontier models cost the same on competing clouds, so the choice there is about region, terms and tooling, not price. And keeping data in one region has a price: pinning Claude to the EU or US adds 10%, and pinning GPT-6.1 Sol to the EU on Azure adds 20%.

How to choose

• Start from the models, not the cloud: list the two or three models your product needs and check which platforms serve all of them.
• Price the same model on each platform: list prices differ by up to 2x for the same open model.
• Check where the data goes: if you need EU residency, count the EU-pinned routes for your models, not just the platform's EU regions.
• Plan for change: models retire and prices move every quarter, so keep the model and the provider as a setting, not as code, which a multi-cloud API such as Eden AI does for you.

Why Eden AI is the best choice: build on all four

Most teams do not need to pick one of these platforms. They need the right model for each task, at the right price, in the right region, and that is spread across all four. Eden AI is a European AI gateway that puts Bedrock, Azure AI Foundry, Vertex AI and Model Studio behind one API, along with around 1,000 AI models from many other providers.

Because the provider is a field in the request, switching cloud does not mean switching code. In this example, GPT-6.1 Sol is served by Azure first, and by Bedrock if Azure is unavailable:

from openai import OpenAI

client = OpenAI(api_key="YOUR_EDENAI_API_KEY", base_url="https://api.edenai.run/v3")

response = client.chat.completions.create(
    model="gpt-6.1-sol",
    messages=[{"role": "user", "content": "Summarise this contract in five bullet points."}],
    extra_body={"routing": {"order": ["azure", "amazon"]}},
)
print(response.choices[0].message.content)

We tested it on 29 October 2026: with order set to Azure first, Azure served the request; with Bedrock first, Bedrock served it. The same call sent to the EU endpoint, https://api.eu.edenai.run/v3, went to Azure's EU deployment, without changing anything else.

What this gives you in practice:

• One key and one bill for four clouds, without four accounts, four contracts and four quota requests.
• Fallbacks across clouds: if one platform hits a limit or an outage, the next one in your routing order takes over.
• The cheapest route for each model: routing: {"sort": "cost"} sends each request to the cheapest provider serving that model.
• EU-only traffic when you need it: the EU endpoint only routes to providers and models cleared for EU data residency.
• Freedom to move: when a new model ships on a platform you do not use, it is one model string away.

Going direct mainly makes sense when your whole stack and committed spend live in one cloud and that cloud serves every model you need. For everyone else, choosing all four through Eden AI beats choosing one.

FAQ

Which is the best LLM platform in 2026?

It depends on the models you need. Azure AI Foundry is the historical home of OpenAI's GPT, Vertex AI is the only one with Gemini, Bedrock has the widest third-party catalogue with Claude, and Model Studio has Qwen and low list prices on open models. If you need models from several of them, a multi-cloud API such as Eden AI gives you all four behind one key.

Which platform has GPT, Claude, Gemini and Qwen?

GPT is on Azure AI Foundry and Bedrock, and in private preview on Google Cloud. Claude is on Bedrock and Vertex AI. Gemini is only on Vertex AI. Qwen is on Model Studio, and some Qwen models are also on Bedrock.

Which platform is the cheapest?

None is cheapest for every model. On 29 October 2026, DeepSeek V4.1 Flash cost half as much on Model Studio as on Vertex AI, while gpt-oss-120b cost 40% less on Vertex AI than on Bedrock. Frontier models such as Claude Opus 5.5 and GPT-6.1 Sol cost the same on competing clouds.

Can I keep my data in the EU?

Yes, with EU-pinned deployments, which usually cost 10 to 20% more. On Eden AI, the EU endpoint only routes to providers and models cleared for EU data residency, across all four clouds.

Can I use several LLM platforms at once?

Yes. With Eden AI, you call Bedrock, Azure AI Foundry, Vertex AI and Model Studio through one OpenAI-compatible API and one key, and switch or fall back between them by changing a field in the request.