Provider Clouds

nRouter routes requests across major enterprise cloud and AI provider networks behind a single unified API key. A provider cloud is a source of models, not a separate integration: models from every cloud below sit in the same catalog, are called with the same request shape, settle at the exact list price, and sit under the same budgets, guardrails, and logs.

OpenAI

Example models: gpt-5.5, gpt-4o, gpt-4o-mini, o1, o3. Capabilities: text generation, structured outputs, tool use, vision, and embeddings — the broadest single-cloud spread in the catalog, and the shape nRouter's OpenAI-compatible API follows most closely.

Anthropic

Example models: claude-sonnet-4-5-20250929, claude-opus-4-20250514, claude-3-5-haiku. Capabilities: reasoning, code analysis, long context, and prompt caching. Anthropic models are reachable through both the Chat Completions endpoint and the native Messages endpoint, with its content blocks, thinking mode, and tool use.

AWS Bedrock

Example models: Claude models, Llama 3.3, Amazon Nova, and DeepSeek. Capabilities: enterprise multi-cloud resilience and cross-region inference. Bedrock is the cloud enterprises most often already standardised on — through nRouter its models need no AWS account or Bedrock credentials of your own.

Google Vertex AI

Example models: gemini-2.5-pro, gemini-2.5-flash, gemini-2.0-flash. Capabilities: multimodal comprehension, audio understanding, and 2M context windows — the catalog's long-context and multimodal depth runs through Vertex AI.

Azure Foundry

Example models: GPT-4o, Phi-4, Mistral Large, and Cohere Embed. Capabilities: direct enterprise cloud egress and managed SLA endpoints, for organisations whose compliance posture already runs through Azure.

Alibaba US

Example models: Qwen 2.5, DeepSeek R1, and DeepSeek V3. Capabilities: open-weight, high-efficiency reasoning and code generation — the route to the Qwen and DeepSeek families under the same key as everything else.

What is the same on every cloud

• One key. Your sk-nrouter- key is the only credential. There are no per-cloud accounts or keys to manage, and nRouter does not do BYOK — which is what lets the controls below apply uniformly.
• One request shape. Copy a model ID from the catalog into a normal Chat Completions request. No cloud has its own endpoint or request fields on nRouter's side.
• One price rule. Every model settles at its exact list price with 0% markup, whichever cloud hosts it.
• One set of controls. Budgets, guardrails, and the fallbacks you configure apply to every request before it leaves for any cloud, and every call lands in your request logs naming the model that served and its exact cost. Analytics can group that traffic by provider, so one cloud's usage is visible on its own whenever you want it.

The live inventory changes as providers release models — the Models page and the models reference are always the current list.

Related guides

• Use Reasoning Models — The reasoning families several of these clouds host
• Send Image Input — Vision input for the multimodal models above
• Stream Responses — Token-by-token answers from any cloud

Frequently asked questions