Model platforms, in the vendors’ own words
Every cell quotes the vendor. Checked 18 Nov 2026: 72 of 72 quotes still on their pages. How
Model platforms, feature by feature
| MCP servers | |||||
|---|---|---|---|---|---|
| Claude models | |||||
| OpenAI GPT models | |||||
| Gemini models | |||||
| Open-weight models | |||||
| Fine-tuning | |||||
| Reserved capacity | |||||
| Content filters | |||||
| Audit logs | |||||
| Data residency | |||||
| Your own encryption keys | |||||
| Private network | |||||
| Not used for training | |||||
| SOC 2 and ISO 27001 | |||||
| HIPAA BAA | |||||
| FedRAMP | |||||
| Copyright protection |
- Yes
- Partly
- No
- Not documented
- Words in cells are each vendor’s own name for the feature. Click a cell for its source.
Can it connect to tools through the Model Context Protocol (MCP)? AgentCore Gateway turns APIs and Lambda functions into MCP tools that agents can call.
it converts APIs, Lambda functions, and existing services into Model Context Protocol (MCP)-compatible tools
Can it connect to tools through the Model Context Protocol (MCP)? The Messages API calls remote MCP servers directly; the MCP connector is in beta.
Connect to remote MCP servers directly from the Messages API without an MCP client, and allowlist, denylist, or configure individual tools.
Can it connect to tools through the Model Context Protocol (MCP)? Agents register and use remote MCP servers through Agent Registry and Agent Gateway.
It provides a unified catalog that lets you securely store, discover, and manage Model Context Protocol (MCP) servers, tools, standalone skills (Preview), and AI agents across your organization.
Can it connect to tools through the Model Context Protocol (MCP)? Foundry agents connect to remote MCP servers, with optional approval of each tool call.
Connect your Foundry agents to Model Context Protocol (MCP) servers by using the MCP tool.
Can it connect to tools through the Model Context Protocol (MCP)? The Responses API calls remote MCP servers; Secure MCP Tunnel reaches private ones.
Remote MCP servers can be any server on the public Internet that implements a remote Model Context Protocol (MCP) server.
Can you call Anthropic’s Claude models here? Current Claude Opus, Sonnet, Haiku, Fable and Mythos models are listed.
The following Anthropic models are available in Amazon Bedrock
Can you call Anthropic’s Claude models here? Anthropic's own Claude models and Claude Managed Agents.
The Claude API is a RESTful API at https://api.anthropic.com that provides programmatic access to Claude models and Claude Managed Agents.
Can you call Anthropic’s Claude models here? Claude models run as a managed API; pay as you go or Provisioned Throughput.
The following models are available from Anthropic to use in Gemini Enterprise Agent Platform.
Can you call Anthropic’s Claude models here? Bought through Azure Marketplace; needs a paid subscription in a country Anthropic sells to.
Anthropic's Claude models bring advanced conversational AI capabilities to Microsoft Foundry, enabling you to build intelligent applications with state-of-the-art language understanding and generation.
Can you call Anthropic’s Claude models here? Searched OpenAI's models and pricing docs; the model list names only OpenAI models.
No statement from OpenAI on its own pages. Searched OpenAI's models and pricing docs; the model list names only OpenAI models.
Can you call OpenAI’s GPT models here? GPT-5.5 and GPT-5.4 went GA in June 2026; GPT-6 models are now listed too.
You can now use GPT-5.5 and GPT-5.4 in production workloads on Amazon Bedrock and build with Codex for AI-powered software development
Can you call OpenAI’s GPT models here? Searched Claude Platform model docs; Anthropic's model list names only Claude models.
No statement from Anthropic on its own pages. Searched Claude Platform model docs; Anthropic's model list names only Claude models.
Can you call OpenAI’s GPT models here? GPT 6 models are in private preview; open-weight gpt-oss-120b is a managed API.
GPT 6.0 Astra is the highest-capability partner model from OpenAI for complex multi-step reasoning, difficult engineering tasks, and long-horizon agent workflows.
Can you call OpenAI’s GPT models here? GPT-6, GPT-5 and gpt-oss models run as Azure OpenAI in Foundry Models.
They include all Azure OpenAI models and specific, selected models from top providers
Can you call OpenAI’s GPT models here? OpenAI's own GPT-6 and GPT-5 models, on the Responses API.
The OpenAI API Platform gives developers access to powerful models like our GPT-5 models.
Can you call Google’s Gemini models here? Searched Bedrock pricing and model pages; Google's open Gemma models are listed, no Gemini model.
No statement from AWS on its own pages. Searched Bedrock pricing and model pages; Google's open Gemma models are listed, no Gemini model.
Can you call Google’s Gemini models here? Searched Claude Platform model docs; Anthropic's model list names only Claude models.
No statement from Anthropic on its own pages. Searched Claude Platform model docs; Anthropic's model list names only Claude models.
Can you call Google’s Gemini models here? Google's own Gemini models are the platform's flagship models.
Flagship models developed by Google, including Gemini, Gemma, Veo, and Lyria.
Can you call Google’s Gemini models here? Searched Foundry model catalog docs; providers listed are OpenAI, Anthropic, Meta, Mistral, DeepSeek, xAI and others, no Gemini.
No statement from Microsoft on its own pages. Searched Foundry model catalog docs; providers listed are OpenAI, Anthropic, Meta, Mistral, DeepSeek, xAI and others, no Gemini.
Can you call Google’s Gemini models here? Searched OpenAI's models and pricing docs; the model list names only OpenAI models.
No statement from OpenAI on its own pages. Searched OpenAI's models and pricing docs; the model list names only OpenAI models.
Can you call open-weight models such as Llama, Mistral or gpt-oss? Llama, Mistral, gpt-oss, DeepSeek, Qwen and Gemma models are offered.
The following Meta models are available in Amazon Bedrock
Can you call open-weight models such as Llama, Mistral or gpt-oss? Searched Claude Platform model docs; Anthropic's model list names only Claude models.
No statement from Anthropic on its own pages. Searched Claude Platform model docs; Anthropic's model list names only Claude models.
Can you call open-weight models such as Llama, Mistral or gpt-oss? Open-weight models run as managed APIs or deploy to your own endpoints from Model Garden.
Support for a wide range of open-weights models like DeepSeek, Llama, and Qwen.
Can you call open-weight models such as Llama, Mistral or gpt-oss? Llama, Mistral, DeepSeek and gpt-oss are sold directly by Azure.
Use the tabs at the top of this page to switch between Azure OpenAI models and Other model collections from providers like Cohere, DeepSeek, Meta, Mistral AI, and xAI.
Can you call open-weight models such as Llama, Mistral or gpt-oss? OpenAI's gpt-oss page offers a Hugging Face download and lists zero API rate limits; no statement it is served.
No statement from OpenAI on its own pages. OpenAI's gpt-oss page offers a Hugging Face download and lists zero API rate limits; no statement it is served.
Can you further train a model on your own examples? Supervised and reinforcement fine-tuning and distillation, for selected models.
You can customize Amazon Bedrock foundation models to improve their performance and create a better customer experience.
Can you further train a model on your own examples? Searched Claude Platform docs and pricing; no fine-tuning offered on the direct API.
No statement from Anthropic on its own pages. Searched Claude Platform docs and pricing; no fine-tuning offered on the direct API.
Can you further train a model on your own examples? Supervised, preference and reinforcement tuning for Gemini; open models can be tuned too.
Model tuning works by providing a model with a training dataset that contains a set of examples of specific downstream tasks.
Can you further train a model on your own examples? Supervised fine-tuning jobs for supported models and training regions.
Prepare your data and create a supervised fine-tuning (SFT) job with Managed Training in Microsoft Foundry.
Can you further train a model on your own examples? Closed to new organizations since May 7, 2026; existing users lose new jobs on Jan 6, 2027.
Creating fine-tuning jobs or training is not available to organizations that have not previously run fine-tuning.
Can you buy guaranteed throughput for steady, high-volume use? Reserve tokens-per-minute capacity for 1 or 3 months; Provisioned Throughput is also sold.
The Reserved tier provides the ability to reserve prioritized compute capacity for your mission-critical applications that cannot tolerate any downtime.
Can you buy guaranteed throughput for steady, high-volume use? Priority Tier is closed to new buyers; Anthropic says to contact sales for guaranteed capacity.
Priority Tier capacity commitments are no longer available for purchase.
Can you buy guaranteed throughput for steady, high-volume use? Fixed-cost subscription in 1-week to 1-year terms that reserves throughput per model.
Provisioned Throughput is a fixed-cost, fixed-term subscription available in several term-lengths that reserves throughput for supported generative AI models on Gemini Enterprise Agent Platform.
Can you buy guaranteed throughput for steady, high-volume use? Provisioned deployments hold dedicated capacity for you, bought in PTUs.
Provisioned throughput is a deployment type in Microsoft Foundry that provides dedicated model processing throughput for your deployment.
Can you buy guaranteed throughput for steady, high-volume use? Prepaid capacity per model for GPT-5.6 and later; Scale Tier covers older models.
Reserved Tier lets you purchase provisioned throughput for a specific model.
Does the platform offer filters you can set for harmful content or sensitive data? Configurable filters for harmful content, denied topics and personal data (PII).
Amazon Bedrock Guardrails offers a consistent user experience to help detect and filter undesirable content and protect sensitive information that might be present in user inputs or model responses
Does the platform offer filters you can set for harmful content or sensitive data? Searched Claude Platform docs; the content moderation guide uses Claude as the moderator, no setting to filter.
No statement from Anthropic on its own pages. Searched Claude Platform docs; the content moderation guide uses Claude as the moderator, no setting to filter.
Does the platform offer filters you can set for harmful content or sensitive data? Configurable safety filters block harmful responses by category and threshold.
To further enhance safety and minimize misuse, you can configure content filters to block potentially harmful responses.
Does the platform offer filters you can set for harmful content or sensitive data? Azure AI Content Safety classifiers check prompts, tool calls and outputs.
Microsoft Foundry provides safety and security guardrails that you can apply to core models and agents.
Does the platform offer filters you can set for harmful content or sensitive data? A free moderation endpoint scores harmful content; your app decides what to block. No PII filter stated.
Use OpenAI moderation models to detect harmful content in text and images.
Can admins export or query a log of who did what? CloudTrail records every Bedrock API call and who made it.
Amazon Bedrock is integrated with AWS CloudTrail, a service that provides a record of actions taken by a user, role, or an AWS service in Amazon Bedrock.
Can admins export or query a log of who did what? The Compliance API's Activity Feed keeps admin and platform activity for 6 years.
The Activity Feed records authentication, chat, file, project, administrative, and platform activity across your organization and returns it in reverse chronological order.
Can admins export or query a log of who did what? Cloud Audit Logs record admin and data access on models and endpoints.
To get audit logs about the usage of your generative models and model endpoints, enable Data Access audit logs for Gemini Enterprise Agent Platform.
Can admins export or query a log of who did what? The activity log records actions on each resource; resource logs go to Azure Monitor.
The activity log contains subscription-level events that track operations for each Azure resource as seen from outside that resource
Can admins export or query a log of who did what? The Admin API's Audit Logs endpoint lists user actions and configuration changes.
Use the Audit Logs endpoint to list recent user actions and configuration changes for the organization.
Can the company choose the region where its data is stored? Stored content stays in the Region you use; cross-region inference can process elsewhere.
Any customer content processed by Amazon Bedrock is encrypted and stored at rest in the AWS Region where you are using Amazon Bedrock.
Can the company choose the region where its data is stored? Stored data can only be kept in the US; US-only inference costs 1.1x. No other regions.
Currently, "us" is the only available workspace geo.
Can the company choose the region where its data is stored? Stored data stays in the chosen location; the global endpoint gives no processing guarantee.
When you store data on Agent Platform (such as custom model weights or metadata), it remains physically stored in the specific Google Cloud location you chose.
Can the company choose the region where its data is stored? Stored data stays in the resource's geography; Global deployments can process elsewhere.
Is stored at rest in the Foundry resource in the customer's Azure tenant, within the same geography as the resource
Can the company choose the region where its data is stored? Per project, for eligible organizations only; 10% price uplift on models released since March 5, 2026.
Data residency is a project configuration option that lets you choose the location of infrastructure OpenAI uses to provide services.
Can the company encrypt its data with keys it controls (customer-managed keys)? Your KMS keys can encrypt custom models, agents, knowledge bases and evaluation jobs.
Use the AWS Key Management Service to manage the keys that you use to encrypt your resources.
Can the company encrypt its data with keys it controls (customer-managed keys)? Keys in your AWS KMS, Google Cloud KMS or Azure Key Vault encrypt certain stored data.
With customer-managed encryption keys (CMEK), you create a key in your own AWS KMS, Google Cloud KMS, or Azure Key Vault.
Can the company encrypt its data with keys it controls (customer-managed keys)? Customer-managed keys in Cloud KMS; Google says security controls don't cover preview models.
If you want to control your encryption keys, then you can use customer-managed encryption keys (CMEKs) in Cloud KMS with CMEK-integrated services including Agent Platform.
Can the company encrypt its data with keys it controls (customer-managed keys)? CMK in Azure Key Vault or Managed HSM encrypts data at rest in the Foundry resource.
This article shows how to configure customer-managed key (CMK) encryption for a Microsoft Foundry resource by using Azure Key Vault or Azure Managed HSM.
Can the company encrypt its data with keys it controls (customer-managed keys)? Bring your own key from AWS KMS, Google Cloud or Azure Key Vault; some endpoints excluded.
Enterprise Key Management (EKM) allows you to encrypt your customer content at OpenAI using keys managed by your own external Key Management System (KMS).
Can apps reach the service over a private network link instead of the public internet? Interface VPC endpoints keep Bedrock traffic off the public internet.
You can use AWS PrivateLink to create a private connection between your VPC and Amazon Bedrock.
Can apps reach the service over a private network link instead of the public internet? Searched Claude Platform docs; found IP allowlisting (beta) but no private link for the direct API.
No statement from Anthropic on its own pages. Searched Claude Platform docs; found IP allowlisting (beta) but no private link for the direct API.
Can apps reach the service over a private network link instead of the public internet? Private Service Connect over Cloud VPN or Interconnect keeps traffic private.
On-premises hosts can reach Agent Platform either through the public internet or privately through a hybrid networking architecture that uses Private Service Connect (PSC) over Cloud VPN or Cloud Interconnect.
Can apps reach the service over a private network link instead of the public internet? Private endpoints (Azure Private Link) connect to Foundry accounts and projects.
This article describes how to establish a private connection to your Foundry account and projects using a private endpoint.
Can apps reach the service over a private network link instead of the public internet? Azure workloads only, through Azure Private Link; not self-service, request it through sales.
OpenAI Private Link lets Azure workloads reach regional OpenAI API endpoints through Azure Private Link instead of connecting directly to public API endpoints.
Is business data excluded from model training by default? Neither AWS nor model providers train on Bedrock inputs or outputs.
No, AWS and the third-party model providers will not use any inputs to or outputs from Amazon Bedrock to train Amazon Nova, Amazon Titan, or any third-party models.
Is business data excluded from model training by default? Inputs and outputs are not used for training unless you send feedback or opt in.
By default, we will not use your inputs or outputs from our commercial products (e.g. Claude for Work, Anthropic API, Claude Gov, etc.) to train our models.
Is business data excluded from model training by default? Google won't train on customer data without permission, for GA and preview models.
Google won't use your data to train or fine-tune any AI/ML models without your prior permission or instruction.
Is business data excluded from model training by default? Applies to models sold by Azure, including Azure OpenAI.
prompts and completions are not used to train, retrain, or improve the base models
Is business data excluded from model training by default? API data is not used for training unless you opt in.
As of March 1, 2023, data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us).
Does the vendor state SOC 2 Type II and ISO 27001 for this product? In scope of AWS SOC 1, 2 and 3 reports and ISO 27001; reports are in AWS Artifact.
Amazon Bedrock is one of the AWS services under ISO Compliance for the ISO 9001, ISO 27001, ISO 27017, ISO 27018, ISO 27701, ISO 22301, and ISO 20000 standards.
Does the vendor state SOC 2 Type II and ISO 27001 for this product? Anthropic also states ISO 27001; reports and certificates are in its Trust Portal.
Security documentation, including SOC 2 Type II reports and ISO 27001 and ISO 42001 certificates, is available on the Trust Portal.
Does the vendor state SOC 2 Type II and ISO 27001 for this product? Listed in scope on Google Cloud's ISO 27001 page and its SOC 2 Type II page.
Generative AI on Gemini Enterprise Agent Platform (formerly Generative AI on Vertex AI)
Does the vendor state SOC 2 Type II and ISO 27001 for this product? Azure SOC 2 and ISO 27001 pages point to a PDF scope list; no readable Microsoft page names Foundry.
No statement from Microsoft on its own pages. Azure SOC 2 and ISO 27001 pages point to a PDF scope list; no readable Microsoft page names Foundry.
Does the vendor state SOC 2 Type II and ISO 27001 for this product? OpenAI also states a SOC 2 Type 2 examination covering the API.
OpenAI maintains ISO/IEC 27001:2022 and ISO/IEC 27701:2019 certifications for the information security and privacy management systems supporting the OpenAI API, ChatGPT Enterprise, and ChatGPT Edu services.
Will the vendor sign a HIPAA business associate agreement that covers this product? Bedrock is a HIPAA eligible service under the AWS business associate agreement.
Amazon Bedrock is in scope for common compliance standards such as Service and Organization Control (SOC), International Organization for Standardization (ISO), is Health Insurance Portability and Accountability Act (HIPAA) eligible
Will the vendor sign a HIPAA business associate agreement that covers this product? BAA signed in the Claude Console; covers eligible API features, beta features generally excluded.
With a signed BAA and a HIPAA-enabled organization, you can use supported API features to process PHI while supporting your organization’s HIPAA compliance.
Will the vendor sign a HIPAA business associate agreement that covers this product? Gemini Enterprise Agent Platform and Generative AI on Agent Platform are listed under the BAA.
The Google Cloud BAA includes Google Cloud's entire infrastructure (all regions, all zones, all network paths, all points of presence), and the following products:
Will the vendor sign a HIPAA business associate agreement that covers this product? Microsoft signs a HIPAA BAA for in-scope Azure services, but the scope list is a PDF; no readable page names Foundry.
No statement from Microsoft on its own pages. Microsoft signs a HIPAA BAA for in-scope Azure services, but the scope list is a PDF; no readable page names Foundry.
Will the vendor sign a HIPAA business associate agreement that covers this product? OpenAI signs a BAA for the API Platform on request.
We are able to sign Business Associate Agreements (BAA) in support of customers’ compliance with the Health Insurance Portability and Accountability Act (HIPAA).
Is this product FedRAMP authorized for US government use? FedRAMP Moderate in commercial Regions; GovCloud lists FedRAMP High status per model.
Amazon Bedrock is in scope for common compliance standards such as Fedramp Moderate, Service and Organization Control (SOC), International Organization for Standardization (ISO)
Is this product FedRAMP authorized for US government use? Not authorized; Anthropic points to Claude on Bedrock or Google Cloud for FedRAMP High.
The Claude API from Anthropic (including Claude Platform on AWS) is not FedRAMP authorized today, so use Bedrock or Vertex when you need a FedRAMP environment.
Is this product FedRAMP authorized for US government use? FedRAMP High for Gemini and Anthropic models; open models are the customer's responsibility.
Consequently, using FedRAMP High authorized Agent Platform services allows for interaction with these supported models in a FedRAMP High environment.
Is this product FedRAMP authorized for US government use? Stated for Azure OpenAI (FedRAMP High); not stated for partner models such as Claude.
Azure OpenAI Service is now approved as a service within the FedRAMP High Authorization for Azure Government.
Is this product FedRAMP authorized for US government use? FedRAMP 20x Moderate since April 2026; OpenAI charges 10% more on FedRAMP endpoints.
OpenAI’s Security and Engineering teams worked through KSI implementation, evidence collection, validation, review cycles, and assessment materials to bring ChatGPT Enterprise and API Platform through the 20x Moderate path.
Does the vendor defend customers if AI output is claimed to infringe copyright? Uncapped copyright indemnity covers Amazon's own models, not third-party models like Claude.
AWS offers an uncapped intellectual property (IP) indemnity for copyright claims arising from generative output of the following generally available Amazon generative AI services: Amazon models
Does the vendor defend customers if AI output is claimed to infringe copyright? Commercial Terms: Anthropic defends paying customers against IP claims over outputs.
alleging that Customer’s paid use of the Services (which includes data Anthropic has used to train a model that is part of the Services) in accordance with these Terms or Outputs generated through such authorized use violates any third-party intellectual property right
Does the vendor defend customers if AI output is claimed to infringe copyright? Covers GA Google models (Gemini, Imagen, Veo), not partner models like Claude or GPT.
Gemini Enterprise Agent Platform API (formerly Vertex AI API) used with generally available versions of these foundation models
Does the vendor defend customers if AI output is claimed to infringe copyright? Covers Azure OpenAI output if required mitigations are in place; partner models not named.
The Customer Copyright Commitment ("CCC") is a provision in the Microsoft Product Terms that describes Microsoft's obligation to defend customers against certain third-party intellectual property claims relating to Output Content.
Does the vendor defend customers if AI output is claimed to infringe copyright? OpenAI defends API customers against copyright claims on generally available features.
we will now step in and defend our customers, and pay the costs incurred, if you face legal claims around copyright infringement
Latest change, 1 Jun 2026: Amazon Bedrock. OpenAI GPT-5.5 and GPT-5.4 generally available on Bedrock. All changes
Who sells what
A few companies sell in most of these categories, so start from the row you already pay for. Everyone else sells one kind.