Claude API from Iran

/v1/chat/completions

OpenAI wire format. One client and one body shape for OpenAI, Anthropic, Google and DeepSeek, so swapping provider is a change to the model string. Best when you want to A/B models or keep an existing OpenAI codebase.

/anthropic/v1/messages

The official Messages API, passed through untouched. Use it when you need cache_control, extended thinking, citations or server-side tools, all four of which the OpenAI compatibility layer silently strips.

Auth

Your 1xai- key on both routes. Authorization: Bearer on /v1; x-api-key plus anthropic-version on /anthropic, exactly as Anthropic requires. No Anthropic account of your own.

Billing

Identical on both: provider list price plus a flat 20% markup, deducted from your Toman wallet. Token usage, including cache-read and cache-creation counts, comes from the provider's own response.

Any model name beginning with claude- routes to Anthropic. Nothing else about the request changes, which is why this route is the cheap way to put Claude behind an interface that already speaks OpenAI.

from openai import OpenAI          # yes, the OpenAI SDK

client = OpenAI(base_url="https://1xai.ir/v1", api_key="1xai-...")

resp = client.chat.completions.create(
    model="claude-haiku-4-5",         # any claude-* name routes to Anthropic
    messages=[{"role": "user", "content": "Summarise this changelog."}],
)
print(resp.choices[0].message.content)

The Claude models exposed here include claude-opus-4-8, claude-sonnet-5 and claude-haiku-4-5; the live catalogue with Toman prices is on the models page. claude-haiku-4-5 is also one of the three models covered by the free chat plan.

The OpenAI schema has no field for a cache breakpoint, a thinking budget or a citation block, so a compatibility layer has nowhere to put them: they are dropped before the request leaves. If your prompt carries a large fixed system block, that silence is expensive. The native route forwards the Anthropic body verbatim.

# pip install anthropic
from anthropic import Anthropic

client = Anthropic(
    base_url="https://1xai.ir/anthropic",  # native passthrough
    api_key="1xai-...",                    # sent as x-api-key
)

resp = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=2048,
    thinking={"type": "enabled", "budget_tokens": 1024},
    system=[{
        "type": "text",
        "text": "Long domain handbook repeated on every call ...",
        "cache_control": {"type": "ephemeral"},   # dropped by the /v1 route
    }],
    messages=[{"role": "user", "content": "Answer using the handbook only."}],
)
print(resp.content[-1].text)
print(resp.usage.cache_read_input_tokens)

The same thing without an SDK. Note the two required headers:

curl https://1xai.ir/anthropic/v1/messages \
  -H "x-api-key: 1xai-..." \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-haiku-4-5",
    "max_tokens": 512,
    "messages": [{"role": "user", "content": "ping"}]
  }'

Google's native surface is available the same way at https://1xai.ir/gemini/v1beta for cachedContents and grounding. Both are documented in full, in Persian, in the API docs.

API calls are pay-as-you-go from a wallet balance: Anthropic's own USD list price for the model, plus a flat 20% markup, converted to Toman. Wallet top-ups run through Zarinpal with a 100,000-Toman minimum, so no international card is involved anywhere in the flow. A new account also receives one server-drawn gift-wheel spin worth 100,000 to 1,000,000 Toman in credit that never expires and can be spent on API usage, which is enough to run real requests before deciding anything. Full pricing.

Your requests pass through an intermediary on their way to Anthropic, which is a real trade-off and should be weighed like any third-party API hop. Model behaviour is not ours to alter: refusals, safety policy and availability are Anthropic's, unchanged by the route you took to reach them.

1xAi is an independent gateway. It is not affiliated with, endorsed by, or a reseller for OpenAI, Anthropic, Google or DeepSeek. Requests are relayed to those providers' official APIs, so each provider's own usage policies apply to what their models will and will not do, and their model availability is theirs to change. Current reachability is published on the status page.