Haiku is in the catalog too: claude-haiku-5.5 and claude-haiku-4.5, with prices on the models page. They run on the Premium channel, not on Cheap Haiku.
You tried to sign up for the Claude API in 2026 and hit a geo-block, a declined card, or a vague "we cannot verify your account" email. It is not you. Anthropic's console is one of the most closed developer signups around, and this year the rules got tighter again. Still, you can call Claude without an Anthropic account. Below: who gets locked out and why, then a working claude-sonnet-5 call in about five minutes through codexkey.
Who cannot sign up with Anthropic directly
Anthropic's own supported countries list is the authoritative reference. In practice, the people who hit the wall fall into a few clear groups:
- Russia, Belarus, and most of the CIS. A geo-IP block at signup, then a hard card-BIN block at payment. A VPN does not help: the country of the issuing bank still differs from the billing country Anthropic requires.
- Parts of MENA and Central Asia. Iran, Syria, Sudan, Cuba, North Korea, and a few others are closed by sanctions outright. The rest of the region is formally supported, but most local cards get rejected.
- Residents of China and Hong Kong with local cards. The console loads, then billing turns down UnionPay and most local issuers.
- Students and freelancers without an international card. Even in fully supported countries, billing wants a card issued in the same country as the account. Virtual cards from many neobanks fail KYC.
- Companies that cannot clear corporate KYC fast. Moving to the production tier takes business documents, and checking them takes days or weeks.
If you recognise yourself here, the direct console is a dead end, not a queue. Waiting gets you no closer.
What got stricter in 2026
Three things changed.
- Geo-IP is checked at the door. Anthropic now compares your IP, browser locale, and phone country when the account is created, before any payment. If one signal disagrees, the signup quietly bounces.
- Card BIN must match the billing country. In 2024 and 2025 this was enforced loosely. In 2026 it is a hard block: a US-billed account with a non-US card fails at the first top-up.
- KYC comes earlier. Anthropic moved it from the production-tier gate down to the first paid usage. Hobby developers who used to sit in a free sandbox now meet KYC on day one.
This is no complaint about Anthropic's policy. The company has real compliance reasons. But the people outside its supported countries carry the cost, and they have to plan around it.
How codexkey works: a gateway, not a workaround
codexkey sits in front of the official Anthropic API as an OpenAI-compatible gateway. You register with an email, top up a balance, and call the same Claude models you would call directly: Sonnet 4.6, Opus 4.8, and the latest snapshots. gpt-5.6-luna handles cheap, high-volume work. Calls go to a standard /v1/chat/completions or /v1/messages endpoint. Behind it, your request is forwarded to Anthropic on codexkey's account, the response streams back to you, and the cost comes off your balance.
Three steps to a working call:
- Register with an email. No phone number, no KYC for normal usage, no geo-block. You get the API key in the dashboard.
- Top up the balance. We accept crypto (USDT, BTC, TON), cards, and SBP (the Russian Faster Payments System) for users in Russia. The minimum top-up is small, and $5 is enough to try it.
- Change the base URL in your SDK. Use the OpenAI SDK with the chat-completions endpoint, or the Anthropic SDK with a
base_urloverride. Both work.
Code: OpenAI SDK
If your code already uses the OpenAI SDK, moving to Claude is a one-line change. Claude answers on the same /v1/chat/completions schema:
from openai import OpenAI
client = OpenAI(
api_key="sk_cdx_...",
base_url="https://api.codexkey.online/v1",
)
response = client.chat.completions.create(
model="claude-sonnet-5",
messages=[
{"role": "system", "content": "You are a senior Go reviewer."},
{"role": "user", "content": "Review this handler for race conditions."},
],
max_tokens=1024,
)
print(response.choices[0].message.content)
Streaming, tool calls, and JSON mode keep the format you already know from OpenAI.
Code: Anthropic SDK with a base_url override
If the project already runs on the official anthropic SDK and you do not want to rewrite it, swap the base URL. The rest stays as it was: message format, content blocks, tool use, prompt caching.
from anthropic import Anthropic
client = Anthropic(
api_key="sk_cdx_...",
base_url="https://api.codexkey.online",
)
message = client.messages.create(
model="claude-opus-4.6",
max_tokens=2048,
system="You are a careful technical writer.",
messages=[
{"role": "user", "content": "Explain prompt caching in two paragraphs."},
],
)
print(message.content[0].text)
The Node, Go, and Rust SDKs accept the same override. If your SDK has no base_url argument, set the ANTHROPIC_BASE_URL environment variable.
Supported models
| Model | ID | Best for |
|---|---|---|
| Claude Sonnet 5 | claude-sonnet-5 | Production assistants, code agents, the default workhorse |
| Claude Opus 4.8 | claude-opus-4.6 | Hard reasoning, planning, one call in place of a retry loop |
| gpt-5.6-luna | gpt-5.6-luna | Classification and routing at high volume, light chat |
Every model goes to the same upstream Anthropic API, so quality and behaviour match a direct call. Context windows, tool use, vision, and prompt caching are fully supported.
How billing works
codexkey bills 1:1 against Anthropic's published rates and adds a transparent margin shown on the pricing page. Tokens are not padded through a "rounding" step, there is no per-request fee, and no monthly minimum. You see the same per-million-token input and output prices Anthropic publishes, with the margin on its own line.
To size your costs, the Claude API pricing guide covers Sonnet, Opus, and gpt-5.6-luna per token, and shows where prompt caching really pays off.
Legal note: a gateway, not a bypass
codexkey is a reseller and API gateway. It does not sidestep Anthropic's Terms of Service. The service holds a commercial account with Anthropic, pays for every token you use, and forwards requests under that account. What you pay codexkey for is the gateway itself: billing infrastructure, top-ups in several currencies, and SDK compatibility, on top of the Claude usage underneath. You use the models under Anthropic's usage policy, the same one direct customers follow.
That is what separates a durable service from a grey-market endpoint that vanishes in a quarter. If you are building something you plan to maintain, the gateway model is the only one worth trusting.
Bottom line: working code in five minutes
If a geo-block, your card, or KYC stopped you, the path is short: register, top up, swap the base URL. Five minutes, a client in OpenAI or Anthropic format, and you call Sonnet 4.6 the way a direct Anthropic customer does. One thing changes: the bill that would have come from Anthropic comes from codexkey.