Akamai launches inference, agent sandbox betas

Akamai’s strongest pitch is a complete enterprise agent stack: model serving, execution, networking, and security under one provider.

• –Akamai Inference spans models from 8B to 2.8T parameters, including Nemotron 3 Ultra, gpt-oss-120b, and Gemma 4.
• –OpenAI-compatible APIs lower migration costs for existing applications, though model-specific tool-calling support may vary.
• –Agent Sandbox uses nested WebAssembly isolation for generated JavaScript or Python, with deny-by-default networking and enforced resource limits.
• –The open-source Spin foundation improves portability, but Akamai still controls the production deployment layer.
• –Beta availability and undisclosed pricing leave performance, economics, and general availability unanswered.