Pay less for the same models
Frontier and open models from every major lab, through one OpenAI-compatible API. Same requests. Smaller bill.
- $1 in free credits
- Pay only per token
- No monthly commitment
HOW IT WORKS
Excess compute,
routed to you.
Companies sell their spare capacity
Teams with committed compute they aren't using resell it to providers at a discount, often far below list price.
We route each request to the best price
One OpenAI-compatible API sends every call to the provider offering that model for the least right now, and fails over if one stalls.
We verify the model is the real model
We check the suppliers we route to, so the model answering is the one you asked for, not a cheaper stand-in. Look for the verified badge.
Audit LLM calls, compare real costs, and migrate compatible code.
Model pricing
PAST 24 HOURS1M input + 1M output tokens · USD · Gateway estimate includes routing fee
Connecting to model catalogs and market prices…
Models in demand.
Source: market activity · current reported counters, labeled 24h upstream. A rolling window has not been confirmed. Top 0 models shown; the share bar and totals include all 0.
The biggest savings.
Current discounts against the cheapest OpenRouter endpoint.
One typical request per model (its measured input-to-output mix, or 80:1 where unmeasured) at our current quote including the routing fee, against the cheapest OpenRouter endpoint for the same request; input and output rates only on both sides. Models without a current OpenRouter reference are excluded.
Discover more. Spend less.
Models with current discounts will appear here when fresh prices are available.
See how much you save.
Recorded market prices against today’s cheapest OpenRouter endpoint, for a typical request of each model. Higher means more savings.
Higher is better. Each point prices a typical request of the model (its measured input-to-output mix) at the recorded market rates, against today’s cheapest OpenRouter endpoint for the same request, using input and output rates only on both sides; OpenRouter history is not recorded. Negative values mean above the OpenRouter reference. Recorded market prices exclude the routing fee; gaps stay visible. In + out = 1M input + 1M output tokens.
Biggest discounts
Our price vs the cheapest OpenRouter endpoint, for a typical request.
Waiting for qualified market observations.
Intelligence per dollar
GPQA Diamond score ÷ in + out price.
Waiting for qualified market observations.
Models in demand
Market-reported request activity.*
Waiting for qualified market observations.
Intelligence leaders
Published GPQA Diamond observations.
Waiting for qualified market observations.
A new base URL.
A better price.
Keep your OpenAI-compatible client. Choose an exact model ID, connect a key, and let GPU Router find the eligible route.
Read the quickstart →// Your next request
const completion = await client.chat.completions.create({
model: "openai/gpt-6-astra",
messages: [{ role: "user",
content: "Build something great." }]
});* The market feed labels activity as 24h; the rolling window has not been confirmed. Other venues do not expose comparable public model-wide usage, so totals are not combined. Benchmark observations: Sep 6, 2026; external endpoints, not marketplace quality guarantees.