We some of the cheapest, most reliable tokens on the market by utilizing heterogeneous hardware, a custom inference engine, kernel-level optimization, and no resale markup — just tokens, served fast.
If you need bulk inference at competitive rates, you can reach us directly at tokens@united-inference.com.
| Model | Input | Cached Input | Output | Context |
|---|---|---|---|---|
| Moonshot's Kimi K3 | $3.00/M | $0.30/M | $15.00/M | 1M |
| Z.ai's GLM 5.2 | $0.75/M | $0.14/M | $2.25/M | 1M |
Prices are per million tokens. Volume discounts available for sustained throughput commitments hosted by us or in your own VPC. We never train on your data and all of our endpoints are ZDR hosted on US-owned hardware.
All models served on our own custom optimized inference stack with custom load balancing and routing all the way down to fully custom CUDA, HIP, Pallas and NKI kernels.