Choose a Model — Smart, Cheap, Fast | TrustedRouter

Smart · Cheap · Fast — pick two<br>Choose the right model for the job.

Describe your task and how private it has to be. We plot every route on the iron<br>triangle, keep the ones that fit, and recommend the model — then show you the one-line shortcut.

Create key<br>Verify gateway

220+ models & routes

3 tiers Open · ZDR · TEE

0 prompt logs by default

How the triangle works<br>Three axes, one honest tradeoff.

Smart is reasoning quality. Cheap is dollars per million tokens,<br>straight from the live catalog. Fast is throughput — Cerebras-served gpt-oss,<br>MiMo Ultraspeed and GLM-4.7 Flash sit right at the tip.

Pick a privacy floor — Open , Zero-retention (ZDR) , or a<br>hardware Trusted Execution Environment (TEE) with end-to-end encryption — and the<br>board only keeps routes that clear it. Every route is attested at<br>trust.trustedrouter.com.

Stop choosingOpenAI SDK

base_url="https://api.trustedrouter.com/v1"

# let the router pick, per request:<br>model="trustedrouter/auto" # best fit, zero-retention<br>model="trustedrouter/cheap" # cheapest capable, TEE<br>model="trustedrouter/fusion" # fuse open models > frontier

The shortcut<br>The corner you actually want rarely exists in one model.

Frontier-smart, real-time-fast, dirt-cheap and private almost never coexist in a single model.<br>TrustedRouter's auto, cheap and fusion routes dissolve the<br>tradeoff — one OpenAI-shaped endpoint that picks per request and proves it never logged your prompt.<br>Create a key

Questions

How do you decide which model fits?<br>Tell us the task and we estimate the intelligence it needs (simple to frontier), the latency you can tolerate (real-time to overnight), and a privacy floor. We keep only the routes that clear all three, then rank them by the smart/cheap/fast tradeoff you set on the triangle.

What do the privacy tiers mean?<br>Open routes any attested provider. Zero-retention (ZDR) only uses providers that contractually keep nothing. Trusted Execution Environment (TEE) runs in confidential compute with end-to-end encryption, so even the provider can't read your prompt.

Which models are fastest?<br>Cerebras-served gpt-oss, Xiaomi MiMo v2.5 Ultraspeed and GLM-4.7 Flash sit at the fast tip of the triangle — hundreds to thousands of tokens per second.

Do I have to pick one model?<br>No. trustedrouter/auto picks the best fit per request, trustedrouter/cheap takes the cheapest capable route in a TEE, and trustedrouter/fusion combines open models to beat any single frontier model — all through one OpenAI-shaped API.

Sign in

Choose a sign in method.

G Continue with Google

= Continue with GitHub

M Continue with MetaMask