Claude gateway · v1

ONE KEY. EVERY MODEL.

Production access to the full Claude family — Opus, Sonnet, Haiku — behind a single OpenAI-compatible endpoint. Point your existing SDK at us, change one string, ship.

Get a key Read the quickstart
Drop-in

Speaks the OpenAI chat format. Keep your client, swap the base URL and key.

Streamed

Server-sent tokens with heartbeats. Tool calls and JSON mode pass straight through.

Metered

Billed by the token, per model. Watch spend live and cap it per key.

The family

Cut from two labs, set in one stone

One endpoint, four models, two labs. Move between them by changing a single string — no re-plumbing, no second contract.

Opus 4.8

Anthropic

Hardest reasoning, agents, long context

Context
1M tokens
Input
$3.00 / M
Output
$15.00 / M
model: "claude-opus-4.8"

Sonnet 5

Anthropic

The production default — balanced speed and depth

Context
200K tokens
Input
$1.80 / M
Output
$9.00 / M
model: "claude-sonnet-5"

Haiku 4.5

Anthropic

High-volume, low-latency, real-time work

Context
200K tokens
Input
$0.60 / M
Output
$3.00 / M
model: "claude-haiku-4.5"

GPT 5.6

OpenAI

Cross-check or route by task alongside the Claude family

Context
400K tokens
Input
$4.00 / M
Output
$18.00 / M
model: "gpt-5.6"

First call

Three steps to a response

A real, ordered path from zero to a streamed answer.

  1. 01

    Mint a key

    Create an account and generate a scoped key. Set a monthly cap on it if you want a hard ceiling.

    export ONYX_KEY="onx_live_..."
  2. 02

    Point your SDK here

    Any OpenAI-compatible client works. Only the base URL changes.

    from openai import OpenAI

client = OpenAI(
    base_url="https://onyxtech.cloud/v1",
    api_key=os.environ["ONYX_KEY"],
)
  3. 03

    Send a message

    Switch models by editing the string — the rest of the call is identical.

    resp = client.chat.completions.create(
    model="claude-sonnet",
    messages=[{"role": "user",
               "content": "Ship it."}],
    stream=True,
)

Pay as you go

Top up. Spend by the token.

No subscription, no seats. Add credit to your balance and it draws down per request at the per-model rates above. Whatever you don't use stays on your balance.

Starter

$10credit ≈ 900 ₽

Enough to wire up the API and run real traffic before you commit.

  • $10 / 900 ₽ on your balance
  • All four models
  • Live balance meter
  • Community support
Add $10 / 900 ₽

Standard +10% bonus

$50credit ≈ 4 500 ₽

Lands $55 / 4 950 ₽ on your balance — the extra is on us.

  • $55 / 4 950 ₽ on your balance
  • Higher rate limits
  • Priority routing at peak
  • Per-key spend caps
Add $50 / 4 500 ₽

Pro +15% bonus

$200credit ≈ 18 000 ₽

Lands $230 / 20 700 ₽ on your balance for teams shipping at volume.

  • $230 / 20 700 ₽ on your balance
  • Highest rate limits
  • Priority routing, always
  • Email support
Add $200 / 18 000 ₽

Fixed rate: $1 = 90 ₽. Need committed volume or an invoiced plan? Talk to us. Balance never expires; top up any amount, any time.

Under load

Cut for production

99.95%uptime, trailing 90 days
312msmedian time to first token
3regions, auto-routed

Turn the lights on.

One endpoint, one key, four models. Top up once — tokens draw down as you use them. Nothing expires.

Get a key