One request, start to finish
Any client that speaks the OpenAI chat format can call wandersim. Change the base URL, the key and the model name.
curl https://wandersim.lol/v1/chat/completions \
-H "Authorization: Bearer $WANDERSIM_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "glm-5.3-unlocked",
"stream": true,
"messages": [{
"role": "user",
"content": "Write a random walk in Python."
}]
}'
import os
from openai import OpenAI
client = OpenAI(
base_url="https://wandersim.lol/v1",
api_key=os.environ["WANDERSIM_KEY"],
)
prompt = "Write a random walk in Python."
stream = client.chat.completions.create(
model="glm-5.3-unlocked",
stream=True,
messages=[{"role": "user", "content": prompt}],
)
for chunk in stream:
print(chunk.choices[0].delta.content or "", end="")
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://wandersim.lol/v1",
apiKey: process.env.WANDERSIM_KEY,
});
const prompt = "Write a random walk in Python.";
const stream = await client.chat.completions.create({
model: "glm-5.3-unlocked",
stream: true,
messages: [{ role: "user", content: prompt }],
});
for await (const chunk of stream) {
process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}
Here is a random walk on a grid:
import random
STEPS = [(1, 0), (-1, 0), (0, 1), (0, -1)]
def random_walk(n):
x = y = 0
path = [(x, y)]
for _ in range(n):
dx, dy = random.choice(STEPS)
x, y = x + dx, y + dy
path.append((x, y))
return path
It returns n + 1 points, starting at the origin.
The response is simulated in your browser. Nothing is sent.
Models and prices
US dollars per million tokens. Cached input is billed at the lower rate automatically.
| Model | Input | Cached input | Output | Context |
|---|---|---|---|---|
glm-5.3-unlockedUnlockedFor coding agents and long chains of tool calls. |
$1.75 | $0.28 | $4.90 | 1M |
kimi-k3.1-pro-unlockedUnlockedFor planning, reasoning and long coding runs. |
$2.10 | $0.21 | $8.40 | 1M |
deepseek-v4.1-flash-unlockedUnlockedFor fast coding and tool use. |
$0.28 | $0.056 | $1.05 | 1M |
glm-5.3-flash-unlockedUnlockedFor everyday coding and bulk jobs. |
$0.21 | $0.021 | $0.77 | 1M |
deepseek-v4.1-flashStandardFor low-cost, high-volume work. |
$0.021 | $0.004 | $0.168 | 1M |
kimi-k3.1-proStandardFor planning, reasoning and coding. |
$0.56 | $0.35 | $5.25 | 1M |
Three steps to your first request
Make an account
Sign up with your email address.
Load a balance
Add $10 or more by card. Your credit shows once the payment is confirmed.
Point your agent at wandersim
Create a key, add wandersim as an OpenAI-compatible provider and pick a model.
Questions
Something else? Ask support.
What does "unlocked" mean?
Unlocked models run on abliterated weights, so they turn down far fewer requests. They can still refuse, and the terms still apply. Standard models are unmodified.
Which coding agents work?
Any tool that lets you set a custom OpenAI-compatible chat endpoint. Give it the base URL, your key and a model name from the price list. Streaming and tool calls are supported.
Can I call it from my own code?
Yes. Use an OpenAI SDK or plain HTTP, as in the example request above.
Do you keep my prompts or responses?
We do not. Our server holds prompts and responses in memory while a request runs and never saves or logs them. The models are run by an outside inference provider, which handles each request under its own policy, so we cannot promise that nothing is kept anywhere. We keep what billing needs: request ID, model, time, token counts and charges.
How do I pay?
Add any whole-dollar amount from $10 to $500 at a time, by card. There is no subscription, and nothing is billed later.
What if a stream gets cut off?
Before a request starts we hold enough credit to cover it. When it ends we charge from the token counts the model reports and release the rest. If a stream is cut off before the model reports them, we charge an estimate based on what was sent and received.