Best execution for intelligence

Cut your LLM bill 50%.

Grade halves the cost of every request it fills. Same key, same model names. You pay for the answer, not the attempt: failed requests are free.

Reducing LLM costs 50%, one request at a time.

Run a real request and see the bill with Grade.

Model you use today
Model name with Grade
bestex/grade/…

A real request. No account needed.

The answer
Press Try it free to run this prompt through Grade.
per 1M output tokens
Before
—
→
With Grade
—

Every model, next to what the lab charges.

Live prices. Never above list.

output price · per 1M tokens

Half of list, answered now. The exact model while the vault is funded; a model matched to it otherwise.

Loading live prices…
0 models at half price this way. Never above list on any model.See all models →

Live on the gateway.

Real requests and what they cost, as they happen.

Fill book0 requests · 30 days
routepaid vs list
Loading…
Recent fillslast 20
Nothing yet.

Switch in three lines.

.env
- OPENAI_BASE_URL=https://openrouter.ai/api/v1+ OPENAI_BASE_URL=https://bestex.dev/v1- OPENAI_API_KEY=sk-or-v1-…+ OPENAI_API_KEY=sk-bx-…- MODEL=anthropic/claude-fable-5.1+ MODEL=bestex/grade/anthropic/claude-fable-5.1
Works with the OpenAI SDK, Anthropic SDK, Vercel AI SDK and Claude Code. Docs →

Good to know

What is Grade?

One word in the model name. Add it and that request is billed at half of list.

Is it the exact model I name?

When the vault is funded, yes: the vault pays the other half. Otherwise a model matched to the one you name answers. Batch always runs the exact model.

What if a request fails?

It is free. Errors, timeouts and empty answers are never charged.

Can I pay more than list?

No. The list price is a ceiling on every request.

How do I pay?

USDG from your wallet. 1 USDG is $1 of usage. No fee, no subscription.

What does the token do?

Stakers get half price on every model and a daily rebate. Part of Bestex's margin on every paid request is also set aside to buy $BESTEX for stakers. Each buy is listed on the Stake page.

Do you keep my prompts?

No. Only billing data: model asked for, tokens, cost, time.