Azure AI Foundry pulls 1,700+ models; Bedrock, Vertex AI trail in catalog size

Amazon Bedrock is AWS's fully managed generative AI service, launched into general availability in 2023. It hosts Anthropic's Claude, Meta's Llama, Mistral, Cohere, DeepSeek, and Amazon's own Titan and Nova models through a single API regardless of which provider sits behind it. Its structural advantage is depth of AWS integration: IAM roles, VPC networking, CloudTrail logging, and KMS encryption wire in the way they do for any other AWS service, because Bedrock is built as one rather than bolted on afterward. For security teams, a workload that passes an AWS review rarely needs a separate audit just because it now calls a foundation model—the same controls already apply. Bedrock carries just over 100 models and charges per-token on a serverless basis by default, with agent runtimes billed at $0.0895 per vCPU-hour via AgentCore.

Azure AI Foundry is Microsoft's rebrand and expansion of Azure AI Studio and Azure OpenAI Service, rolled into a single portal and SDK. Its defining feature is exclusivity: under Microsoft's commercial partnership with OpenAI, Azure AI Foundry is the only major cloud platform offering first-party access to the full GPT-5 family, including GPT-5, GPT-5 mini, GPT-5 nano, GPT-5.3, GPT-5.4, and GPT-5.5. Foundry also carries the largest overall model catalog of the three, at 1,700-plus models, spanning OpenAI's GPT-5 line, Meta's Llama, Mistral, Microsoft's own Phi, Alibaba's Qwen, DeepSeek, Cohere, and Anthropic's Claude. It plugs directly into Microsoft 365, Teams, SharePoint, and Copilot, which matters most to organizations already standardized on Microsoft's productivity stack. Azure AI Foundry documentation frames the platform as a full production lifecycle tool, covering evaluation, red-teaming, and observability alongside inference. Pricing mirrors that ambition: customers can choose per-token billing or Provisioned Throughput Units (PTUs) for predictable, steady-state enterprise traffic, with no separate runtime fee beyond inference and tool computation costs. GPT-5 input pricing is $1.25 per million tokens and output is $10 per million tokens, at least below the 200K context threshold.

Google Vertex AI is Google Cloud's unified machine learning platform, home to the Gemini model family and Google's open-weight Gemma line, plus a curated Model Garden hosting Claude, Llama, and Mistral. Vertex runs smaller and more curated than Azure, at roughly 200-plus models, and carries the distinction of originating the Agent-to-Agent (A2A) protocol, a Google-led open standard for letting AI agents from different vendors communicate. Vertex leans harder into the full ML lifecycle than the other two, with strong AutoML tooling and tight integration with BigQuery for teams already doing analytics on Google Cloud. It also carries the most academic pedigree of the three—Google's research organization, the group behind the original Transformer architecture, feeds directly into Vertex's tooling roadmap. Vertex is the only platform of the three with a true free tier for development and low-volume production use. Gemini 2.5 Pro input pricing matches Azure's headline rate at $1.25 per million tokens, but output pricing doubles above a 200K context threshold; Gemma 4 26B is available at roughly $0.13 per million tokens. The Agent Development Kit is free, though underlying infrastructure (Cloud Run/GKE) scales separately.

The catalog gap—Azure's 1,700-plus models against Bedrock's 100-plus and Vertex's 200-plus—is the widest metric in the comparison. But OpenAI's GPT-5 exclusivity on Azure and native data integrations to each parent cloud shape the actual decision more than raw model count. The models most production teams reach for—Claude, Llama, Mistral, and DeepSeek—appear on all three platforms, capping lock-in if you standardize on open-weight lines. The real decision hinges on whether your organization is committed to GPT-5 specifically (Azure is the only choice), already standardized on Microsoft productivity tools (Bedrock's AWS integration, Vertex's BigQuery tie), or running experimental or bursty workloads (Vertex's free tier). Each platform's compliance and security posture also follows its origin: Bedrock publishes granular SOC and FedRAMP certifications; Azure frames itself as governance-first; Vertex appeals to teams whose primary job is training and evaluating models.

The three platforms emerged from different corporate starting points and retain those strategic DNA strands. Amazon Bedrock was built as a native AWS service, which means security teams already familiar with AWS IAM, VPC networking, CloudTrail logging, and KMS encryption see Bedrock as just another AWS service—no separate audit needed. Microsoft Azure AI Foundry, conversely, is the rebrand of Azure AI Studio and Azure OpenAI Service rolled into one surface, and it inherits both Microsoft's productivity ecosystem (365, Teams, Copilot, Sentinel) and, crucially, exclusive first-party access to OpenAI's frontier models under the two companies' commercial partnership. Google Vertex AI, the ML research group's platform, retains the strongest data-pipeline integration of the three—BigQuery is native, and the platform originated the Agent-to-Agent (A2A) protocol, an open standard for cross-vendor AI agent communication. Each platform's compliance and pricing structures follow that origin story: Bedrock publishes granular security certifications (SOC 1/2/3, HIPAA, FedRAMP High, ISO 27001, GDPR) as selling points to regulated finance teams; Azure frames itself as a production lifecycle tool with evaluation and red-teaming, appealing to large enterprises that need governance as much as model access; Vertex bets on teams whose primary job is training and evaluating models, not just running hosted inference. The catalog gap—1,700 models on Azure versus 200-plus on Vertex and just over 100 on Bedrock—matters less than it appears on paper: the models most production teams actually need (Claude, Llama, Mistral, DeepSeek) appear on all three. The real lock-in is OpenAI's GPT-5 exclusivity to Azure and the each platform's native integrations to its parent cloud.

No comments yet. Be the first to share your thoughts!