These services host AI models so you do not own any hardware. Three sub-types: multi-model routers (one API key, many providers), managed inference (one provider serving many open models), and the big-cloud model gardens inside AWS, Google Cloud, and Azure.
Pick a router like OpenRouter or Hugging Face's Inference Providers to try many models through one key. Pick managed inference (Together, Groq, Fireworks, Replicate) for speed and price on specific open models. Pick Bedrock, Vertex AI, or Azure AI Foundry when you are already in that cloud and need enterprise controls.
Pricing is per usage almost everywhere, usually per million tokens. Groq is known for very fast generation on its custom hardware, and Hugging Face's router adds free monthly credits with provider-parity pricing after that.
Snapshot of 7 June 2026. 11 entries in this group, about 1 with a free option and 0 that can run on your own hardware. Prices and versions change often, and each name links to its official page. The full directory has search and filters across all categories.
How to Choose
Try many models with one key:
OpenRouter routes to 400+ models across 60+ providers. Hugging Face Inference Providers does the same across its partners, with free monthly credits.
Fastest generation:
Groq's custom LPU hardware is built for token speed on open models.
Media models via API:
Fal and Replicate specialize in image, video, and audio models.
Enterprise and compliance:
Amazon Bedrock, Google Vertex AI, and Azure AI Foundry put frontier and open models behind the security and governance of the cloud you already use.
All 11 Developer access & hubs Entries
| Tool | Maker | What it is | Cost | Where it runs | Hardware |
|---|---|---|---|---|---|
| Amazon Bedrock | AWS | Enterprise catalog of many models with AWS security and compliance. | Paid | Cloud | Any device |
| Azure AI Foundry | Microsoft | Microsoft enterprise model catalog with Azure governance. | Paid | Cloud | Any device |
| DeepInfra | DeepInfra | Serverless inference marketed on low per-token cost for open models. | Paid | Cloud | Any device |
| Fal | Fal | Fast inference platform centered on generative media. | Paid | Cloud | Any device |
| Fireworks AI | Fireworks AI | Managed inference for open and some closed models, with fine-tuning. | Paid | Cloud | Any device |
| Google Vertex AI | Google Cloud | GCP model platform. Gemini plus open and third-party models. | Paid | Cloud | Any device |
| Groq | Groq | Custom LPU hardware for very fast token generation on open models. | Paid | Cloud | Any device |
| Hugging Face Inference | Hugging Face | Unified API on the Hub routing to many partner inference providers. | Free + paid | Cloud | Any device |
| OpenRouter | OpenRouter | One API routing to 400+ models with fallback and billing in one place. | Paid | Cloud | Any device |
| Replicate | Replicate | Run thousands of community and proprietary models via API. | Paid | Cloud | Any device |
| Together AI | Together AI | Managed inference and fine-tuning for a large open-model catalog. | Paid | Cloud | Any device |
Common Questions
What is a model router?
One API that forwards your requests to many different model providers. You hold one key and one bill, and you can switch models without changing code. OpenRouter and Hugging Face Inference Providers are the main examples.
Do I need my own GPU?
No. That is the point of these services: the provider runs the hardware and you pay per usage, usually per million tokens or per second of compute.
Is any of this free?
Hugging Face Inference Providers includes free monthly credits. The rest are pay-per-use, some with free trial credit. For free unlimited use, you run open models on your own hardware instead.
Keep Reading
Verified articles related to this page:
Other model pages: