API

API reference

One API, compatible with OpenAI's chat completions, completions and responses and with Anthropic's messages, serving Qwen3.8 27B Uncensored, a model for coding and chat with 262K context. Point the SDK you already use at the base URL and send your key.

Base URL

https://api.unrestricted.sh/v1

OpenAI clients use this base URL. Anthropic clients use https://api.unrestricted.sh, without /v1.

Authentication

Send your key as Authorization: Bearer <key> or in the x-api-key header. Keys start with unr_sk_ and work after your account's first top-up.

Endpoints

POST /chat/completions: OpenAI chat completions, with tool calls
POST /completions: OpenAI text completions
POST /responses: OpenAI Responses API
POST /messages: Anthropic Messages API, with tool use
POST /messages/count_tokens: Anthropic token counting
GET /models: The live models

Paths are relative to the base URL.

Streaming

Set "stream": true and the reply arrives as server-sent events, in the format of the endpoint you called.

OpenAI SDK

python

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.unrestricted.sh/v1",
    api_key=os.environ["UNRESTRICTED_API_KEY"],
)

resp = client.chat.completions.create(
    model="qwen3.8-27b-uncensored",
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

typescript

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.unrestricted.sh/v1",
  apiKey: process.env.UNRESTRICTED_API_KEY,
});

const resp = await client.chat.completions.create({
  model: "qwen3.8-27b-uncensored",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(resp.choices[0].message.content);

Anthropic SDK

python

import os
from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.unrestricted.sh",
    api_key=os.environ["UNRESTRICTED_API_KEY"],
)

message = client.messages.create(
    model="qwen3.8-27b-uncensored",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)
for block in message.content:
    if block.type == "text":
        print(block.text)

curl

curl https://api.unrestricted.sh/v1/chat/completions \
  -H "Authorization: Bearer $UNRESTRICTED_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-27b-uncensored",
    "messages": [{"role": "user", "content": "Hello"}],
    "stream": true
  }'

Reasoning

In chat completions the reasoning arrives in reasoning_content, apart from the reply. reasoning_effort caps how long the model reasons before it answers:

none: no reasoning
low (or minimal): up to 1,024 reasoning tokens
medium: up to 4,096
high (or xhigh, max): up to 16,384

Without it, reasoning has no cap, and any other value is ignored. On /responses, send it as reasoning.effort. On /messages, thinking.budget_tokens sets the cap and leaving thinking out turns reasoning off.

Limits

Context: 262,144 tokens
Requests at once: Up to 3 requests may run at once on an account, 5 on Pro and 8 on Max, shared by your API keys and the playground. One more is refused with 429 until a running request finishes.
Request time: Up to 5.5 minutes per request
Wake fee: A request that wakes a sleeping GPU adds $0.01. Nothing extra is charged while the model is already running.

Privacy

Prompts and completions are never saved or logged.

We keep token counts for billing and a few account events, such as that credit ran out. Never prompts or replies.

No content filters.

The privacy policy lists everything that is kept and why.

Errors

401 invalid_api_key: Missing or revoked API key.
402 insufficient_funds: Out of funds.
403 api_locked: Add funds once to unlock API keys.
404 model_not_found: Unknown model id.
429 concurrent_request_limit: Too many requests running on this account.
503 model_not_available: Model not available.

Errors use OpenAI's error format. /messages answers in Anthropic's format and reports a low balance as 400. The 429 for too many requests at once carries a retry-after header and costs nothing.

Pricing, the Qwen3.8 27B Uncensored page, and setup guides for coding agents.