// Less time plumbing models, more time building

You've got plenty to build.
Don't burn it all on wiring models.

One endpoint, 19 models. OpenAI-compatible and Anthropic-native; every request reports its cost in the response headers.

19models, switch whenever you like
2OpenAI-compatible · Anthropic-native
10% offThird-party routes, prices shown up front
20 ★Verify your email and it's yours. No card.
terminal⧉
$ curl https://3306.ai/v1/chat/completions \  -H "Content-Type: application/json" \  -H "Authorization: Bearer YOUR_KEY" \  -d '{    "model": "gpt-5.5",    "messages": [{      "role": "user",      "content": "I want to build a reading-notes app. What should v1 include?"    }]  }'
200 OK · Official direct · X-Stars-Cost: 2
Already have a project? Set base_url and your Key for the protocol you use, then send one request to confirm it works.
3306.AI Token Platinum Card · Hand it over, or send it online

They say "I've got an idea."You say "Compute's on me."

A friend with a spark, a partner with a dream;boss, colleague or client — send compute to the whole team.Don't ask if the idea will fly — let a model give it a try.

Friend: that wild idea — try it?Partner: your idea, my computeBoss: new idea, go run itColleague: compute's on me this timeClient: a gift they'll actually use
Gift compute now
Physical card for handing over in person, e-card for sending online. Pick any amount. The code redeems as stars that never expire, ready whenever the idea shows up.

Building the product is enough work. These six things still need handling.

Endpoints to configure, routes that might get swapped, bills you have to guess at, Keys you hand out and then worry about. Here's how we handle each thing that bugs you when wiring up models.

OPENAI_BASE_URL=https://3306.ai/v1

Switch models without rebuilding the integration

Using the OpenAI SDK? Point it at /v1. Using the Anthropic SDK or Claude Code? Use the native /anthropic. Cursor, Cherry Studio and NextChat just need the address and a Key. Send one request and you're in.

route = official | relay | specified

Picked official? We won't quietly reroute you

You pick the route when you create a Key. Official direct calls only the vendor's official API, at list price. Third-party routes pick the cheapest of several upstreams, from 10% off; you can also pin one. Official and third-party never switch into each other.

X-Stars-Cost: 2

Know what a request cost without waiting for the bill

Every response carries its cost in the X-Stars-Cost header, and 1 ★ = $0.01. Cache hits, long context and image sizes are all spelled out in the pricing table; the displayed price and the actual charge use the same formula.

primary failed → next upstream

Reply cut off midway? You don't pay for what you didn't get

If a streamed reply breaks off, you're charged only for what was delivered. When a model has several upstreams, a failed primary fails over to the next, still within the route you chose; circuit breakers and health checks back it up.

image · video · audio · embeddings

An idea is more than a chat

Image input, image generation, video, speech, vector search — whatever the model catalog currently offers, all with the same Key and the same billing. 17 of the 19 models can see images.

key_limits = model + spend + time + ip

Hand out a Key without handing out the budget

When you set up a Key for a project, decide which models it can use, how much it can spend and how long it lasts. Cap spend per 5 hours, per day, per 7 days and in total; restrict IPs and RPM/TPM too. Using Claude Code? Generate the config straight from here.

Get one request through, then wire it into your product.

01

Create an account, your way

Sign up with email, Google, GitHub or a passkey. No card needed. Verify your email and get 20 ★ to start with.

+20 ★Yours once your email is verified
02

Create a Key. Set the route and budget together

Choose official direct, a third-party route, or pin one upstream. Then set models, spend, expiry and IP limits. The Key is shown once, so save it.

API KeySave it this time
03

Send a request and check what it cost

Set base_url and your Key for the OpenAI or Anthropic protocol. When the response comes back, check X-Stars-Cost: is the result right, and what did it cost? Confirm both at once.

X-Stars-CostThis request's cost lives here

Here's the code. Drop in your Key and try it.

curl https://3306.ai/v1/chat/completions \
  -H "Authorization: Bearer sk_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.5",
    "messages": [{"role": "user", "content": "I want to build a reading-notes app. What should v1 include?"}]
  }'

Know where the money goes. Don't guess.

No monthly fee, no daily cap. Stars you top up never expire, so there's no rush to spend them. 1 ★ = $0.01. Every request reports X-Stars-Cost, and the pricing table and the actual charge use the same formula.

See what you'll get before you decide how much.

Just trying it out, or planning to use it for a while? Pick the amount that fits — what you pay is what you get. The per-$1 rate is on the right. Pay with Stripe (incl. WeChat Pay), WeChat Pay, Binance Pay, and USDT on-chain.

Pick a payment method
Top upCredited per $1For example
Any amount$1.00Top up $10, get $10.00
Third-party routes · from 10% off

Same model. No need to shop around yourself.

Choose a third-party route and we pick the lowest price among the available upstreams. Official and discounted prices sit side by side, so the savings are plain to see. The examples below are per 1M input tokens.

gpt-5.5 · per 1M input tokens$5.00$4.5010% off
gpt-5.6-sol · per 1M input tokens$4.00$3.6010% off
gpt-5.6-terra · per 1M input tokens$2.00$2.00
Create a Key on a third-party routeSee prices for all 19 models
Official direct · list price

Going official? Then it goes exactly where you chose.

Some projects need to know exactly where requests go. With official direct, requests reach only the vendor's official API, billed at the vendor's list price, and never switch to a third-party route. Pick it when you create the Key and you know what you're getting.

How much runs at once (by plan)Concurrency 5–80 / RPM 600–6,000
What this request costResponse header X-Stars-Cost
Scope this KeySpend cap · IP allowlist · Expiry
Create a Key on official direct
That little thing you've always wanted to build —
get it running first.
https://api.openai.com/v1→https://3306.ai/v1
No card needed. Verify your email and get 20 ★. Pick a model, set base_url and your Key for the OpenAI or Anthropic protocol, and try the thing you actually want to build.