| Model catalogue | Over 10,000 models, including OpenAI, Cohere, Mistral and Meta Llama, through one endpoint. | 19 models across 6 providers, each priced and benchmarked rather than listed. | Their advantage |
| Anthropic Claude | Not part of the Foundry catalogue. Evaluations that land on Claude need Bedrock or Anthropic directly. | Claude is a first-class destination, and Claude Code can route through the gateway via the Anthropic Messages endpoint. | Our advantage |
| Cloud coupling | Azure-native. You pay for the underlying Azure services consumed — inference, storage, compute, retrieval, monitoring. | Runs independently of any hyperscaler, calling providers directly or a self-hosted endpoint. | Our advantage |
| Governance | Built in: version management, evaluation frameworks, audit documentation, content safety and policy-aligned guardrails. | Policy is the trust boundary, composed by deny-overrides so a team can tighten but never loosen, with every version immutable and recorded per execution. | Comparable |
| Multi-model routing | Advanced multi-model routing is still in public preview. | Deterministic planning per request, measured against a benchmark whose results — including the unflattering ones — are published. | Our advantage |
| Many clients or teams | Expressed through Azure subscriptions, resource groups and RBAC. | The client organisation is the product primitive, with its own policy, budget, statement and client-facing read-only login. | Our advantage |
| Certifications and support | Microsoft's full compliance and enterprise support estate. | None, and limited beta. Said plainly rather than buried. | Their advantage |