Prime Inference
OpenAI-compatible serverless API for open models such as GLM, DeepSeek and Kimi billed per token, plus dedicated deployments of custom or fine-tuned models on reserved hardware with a 99.9% uptime SLA.
Find alternatives to Prime Inference
What it does
- Model inference
- Model hosting
Sources
- Pricing
-
The inference page's FAQ states how it is priced. Page quote: "The Serverless API, Prime Gateway included, is billed per token. Dedicated capacity for your own models is sized and priced with our team.".
- Description
-
Official documentation · 16 Nov 2026
The vendor's own page describes the product. Page quote: "Call frontier open models per token through one OpenAI-compatible API.".
- What it does
-
Official documentation · 16 Nov 2026
Capabilities follow the functions the page names. Page quote: "Serve GLM, DeepSeek, Kimi and more through one API." / "Your own endpoint, sized to your traffic.".
- Sold within
-
Per-token billing with a get-started route; dedicated capacity is a sales conversation. Page quote: "Pay per token, with nothing to deploy or manage.".
- Name
-
Official documentation · 16 Nov 2026
Name as the vendor's own page prints it. Page quote: "Call open models per token, or serve your own on dedicated capacity.".
- URL
-
Official documentation · 16 Nov 2026
Product page text fetched with curl 2026-11-16. Page quote: "Call open models per token, or serve your own on dedicated capacity.".
- Deployment
-
Official documentation · 16 Nov 2026
The vendor runs the endpoints; a dedicated deployment is provisioned by the vendor. Page quote: "For a Dedicated Deployment, Prime provisions the hardware and our team manages the serving.".
- Interfaces
-
Official documentation · 16 Nov 2026
How the vendor's own page says the product is reached. Page quote: "Call frontier open models per token through one OpenAI-compatible API.".
Also from Prime Intellect
-
Model Factory Platform
Prime Intellect's hosted platform that takes a model from task and environment through reinforcement-learning training and hosted evaluation to a live serving endpoint, so teams can train specialist models for their own tasks.
-
Prime Compute Infrastructure service
GPU cloud that rents H100, H200 and B300 GPUs by the hour with no contract, single GPUs up to multi-node clusters, plus dedicated InfiniBand clusters set up with the vendor's team on a quote.
-
Prime Sandboxes Infrastructure service
Linux microVM sandboxes that run tens of thousands of concurrent isolated environments for agentic reinforcement-learning training, billed per vCPU, GiB of memory and GiB of disk per hour and used through a CLI and SDK.
Open Prime Intellect in the directory
Something wrong here? Send a correction — quote this product id: prime-intellect-inference.