OpenRouter compared with LiteLLM, Amazon Bedrock and Azure OpenAI in Microsoft Foundry Models| OpenRouter | LiteLLM | Amazon Bedrock | Azure OpenAI in Microsoft Foundry Models |
|---|
| Tipo de serviço: OpenRouter is a hosted gateway. One endpoint, OpenAI SDK compatible through the base URL, gives access to hundreds of models and handles fallbacks between providers. | Tipo de serviço: LiteLLM is an open source project. It is a Python SDK with a Router for retries and fallbacks, and a self-hosted proxy that speaks the OpenAI format to more than 100 providers. | Tipo de serviço: Bedrock is Amazon's managed platform for generative AI applications and agents. It sits inside an AWS account and adds guardrails, evaluation tools and distillation. | Tipo de serviço: Microsoft now documents these as Foundry Models sold by Azure, covering OpenAI GPT, o-series, gpt-oss, image, audio and embedding models. They are billed through an Azure subscription and covered by Azure SLAs. |
| Model coverage: The Standard plan lists 500+ models from 80+ providers, plus a free tier with 25+ free models. Many models from different labs can be tried without separate registrations. | Model coverage: Coverage depends on the keys you hold. LiteLLM routes to providers including OpenAI, Anthropic, Vertex AI and Bedrock, but does not host models. | Model coverage: Bedrock offers hundreds of foundation models from major AI companies. The overview page does not name the providers. | Model coverage: Mainly OpenAI-family models, with open-weight gpt-oss and a separate catalogue of partner and community models. Availability varies by region and quota tier. |
| Tratamento de dados: Prompts and responses are not stored unless you opt in. You can block routing to providers that may train on your data, and filter providers by data policy. EU in-region routing is reserved for enterprise customers. | Tratamento de dados: Self-hosting keeps the gateway under your control. What each upstream provider keeps is set by that provider and your own key. | Tratamento de dados: The vendor states Bedrock never stores or uses your data to train models. Listed compliance includes ISO, SOC, GDPR, FedRAMP High and HIPAA eligibility. | Tratamento de dados: Data stays within the customer's Azure tenancy and its compliance terms. Check the region availability list before choosing a deployment. |
| Modelo de custo: Pay-as-you-go. Platform fees are listed at 5.5 percent (Standard) and 8 percent (Business), and the free plan allows 50 requests per day. Model prices come on top. | Modelo de custo: The software is MIT licensed with no per-call fee. Enterprise features such as SSO and audit logs are offered separately, and you pay the providers directly. | Modelo de custo: Billed through AWS, with a separate pricing page. Batch processing, prompt caching and routing are offered to cut cost. | Modelo de custo: Billed through Azure per token, with pricing that depends on input size categories and deployment type. |
| Quando escolher: Choose it to compare many models quickly with one account, or to prototype without contracting each provider. | Quando escolher: Choose it when a team wants its own gateway with budgets, virtual keys and cost tracking across providers. | Quando escolher: Choose it when workloads already run on AWS and need its governance and compliance controls. | Quando escolher: Choose it for Microsoft-centred organisations that need OpenAI models under Azure contracts and SLAs. |
