Looking for Akamai Inference Cloud? Explore Akamai Cloud. Looking for Akamai Inference? You’re on the right page.
Overview
Consume open-weight models without vendor lock-in or GPU management through an OpenAI-compatible API on Akamai Cloud
Akamai Inference is a fully managed inference service for running AI models in production. Call them through an OpenAI-compatible API while Akamai handles GPUs, serving, autoscaling, and model lifecycle management. Use pay-per-token Serverless Inference, Dedicated Inference with reserved GPUs, or both.
Why use Akamai Inference?
Get the agility of a modern inference provider without giving up enterprise trust. Build the full AI application, not just the model call. Grow from first experiment to production on one platform.
How it works
Use Akamai Inference to power your intelligence
Fast and easy access to the leading open-weight models. Power your agentic needs while avoiding vendor lock-in and high token costs. Run production inference with predictable performance and a dedicated OpenAI-compatible API endpoint backed by GPU capacity reserved exclusively for you.
Serverless, Dedicated, or both
Match each workload to the right deployment type. Use one or both from the same provider, through the same API.
Features
AI inference features for deploying the latest open-weight models
- Curated catalog of leading open-weight models, including Llama, Qwen, and DeepSeek
- Dedicated Inference on GPU capacity reserved for a single customer
- Audit logging for control plane operations and inference calls (limited availability)
- OpenAI-compatible API for drop-in integration
- Choice of region and GPU configuration for dedicated endpoints
- Chat playground for testing models
- Pay-per-token Serverless Inference with automatic scaling
- Zero data retention for prompts and responses
- API key management
Use cases
Akamai Inference use cases
Run leading open-weight models in production, or deploy your own custom and fine-tuned models with the same security and controls. The inference agility you need, from the provider you already trust.
Pick from the latest leading open-weight models and start with Serverless Inference.
Serverless Inference offers the leading open-weight models on shared GPU pools. You pay on a pay-per-token basis for what you use.
Run a dedicated, OpenAI-compatible model endpoint.
Dedicated Inference allows you to choose the model, region, GPU configuration, and scale. You get a dedicated, OpenAI-compatible model endpoint reserved solely for you.
Frequently Asked Questions (FAQ)
Frequently Asked Questions (FAQ)
Serverless Inference serves models through a shared, autoscaling endpoint and is priced per million tokens. It’s designed for getting started and for bursty or unpredictable workloads. Dedicated Inference provides a private endpoint on GPU capacity reserved for a single customer, in the region and GPU configuration you choose, and is priced per GPU-hour. It’s designed for steady, high-volume production traffic. You can use either one or both. Dedicated Inference will have limited availability.
The model catalog includes leading open-weight models. The catalog will expand, and new models are added as released.
There’s no charge during the beta. When billing begins at limited availability, Serverless Inference will be priced per million input and output tokens. Dedicated Inference will be priced per GPU-hour, billed by the minute for as long as an endpoint is provisioned, including idle time.
No. Akamai Inference uses zero data retention, so prompts and responses are not stored or used to train models.
Yes. Akamai Inference uses an OpenAI-compatible API, so most applications built for the OpenAI API can switch by changing the base URL and API key, and selecting a model from the Akamai Inference catalog. (Support for specific API features, such as tool calling or structured outputs, may vary by model.)
Serverless Inference is a deployment model where you call AI models through a shared, autoscaling endpoint. You pay only for the tokens you consume, with no need to provision, manage, or maintain underlying GPU infrastructure.
An open-weight model is a model whose trained weights are publicly released, so anyone can download, run, and fine-tune it. Examples include Llama, Qwen, and DeepSeek.
Sign up for the Serverless Inference beta
Request access to the Serverless Inference beta.
- Discover fast and simple access to inference service
- Run open-weight models on Akamai Cloud
- Eliminate provisioning and managing GPU infrastructure
- Build latency-sensitive, real-time AI apps and agents
