Cheaper Inference
Cheaper Inference
We’re driving down the price of compute so more people can build with frontier AI models and dedicated GPUs.
Estimated tokens served
Loading usage…
01 / The model marketplace
Access the world’s leading AI models at prices that give you room to go further.
| Model | Input | Output | Discount | Capabilities | |
|---|---|---|---|---|---|
| deepseek-v4.1-flash-1017deepseek-v4.1-flash-101756% offDeepSeek | Input | Output | 56% off | Capabilities | |
| deepseek-v4.1-flashdeepseek-v4.1-flash45% offDeepSeek | Input | Output | 45% off | Capabilities | |
| glm-5.3-flashglm-5.3-flash60% offZ.ai | Input | Output | 60% off | Capabilities | |
| glm-5.3glm-5.345% offZ.ai | Input | Output | 45% off | Capabilities | |
| claude-opus-5.5claude-opus-5.535% offAnthropic | Input | Output | 35% off | Capabilities+1 | |
| glyph-clusterFreeglyph-clusterFreeStealth | InputFree | OutputFree | Capabilities |
Usage based. No separate routing surcharge. Never above direct list price.
02 / A simpler way in
Replace your base URL and API key. Keep your messages, tools, streaming, and response handling exactly where they are.
// Two values to change
baseURL: "https://api.openai.com/v1"
apiKey: not set
03 / Trust, by design
See what the marketplace records, where caching can occur, and which routes support zero-data-retention controls.
Turn on zero data retention and only usage and billing metadata is kept.
Read the privacy policyChoose supported routes when your workload requires zero-retention controls.
Check current security statusEach request reaches only the provider selected to serve that route.
Review subprocessorsCache controls pass through when the selected model supports them.
Review caching guidance04 / Dedicated compute
NVIDIA B300 GPUs are offered on one- and two-year commitments. Explore rates and request a quote for the capacity you need.

One more thing.
The most-starred project in GitHub’s AI gateway topic. An open-source router connecting hundreds of providers, free for anyone to run.
Open source. MIT licensed. Built with the community.
05 / Build the marketplace
Sell inference through an existing API key, or offer your GPUs for rental.
Connect a provider account that permits routed customer requests. We test your key and list the models it serves.
Already a seller? Open the seller dashboard.
Seller Agreement · Privacy Policy
Questions, answered
Pricing, reliability, and data handling, in plain terms. Still unsure? Talk to our team.
The future has room for you
Save up to 100% on the models you already use. Create a free account, fund from $25, and keep your current request format.