Standard API
Open-model inference, billed per token
Call supported models through the Standard API, which is compatible with OpenAI’s Chat Completions format. Input and output tokens are billed separately. Cache-read rates follow each provider’s caching rules.
Choose a model for your workload.
Text, image, and video models behind one base URL and API key. Standard One answers at /v1/systemone instead of /v1/chat/completions; see how to call it. Availability depends on capacity, and a listed model may not be available yet.
Connect your application in three steps.
Set your API key, the Standard Thinking base URL, and a model ID enabled for your workspace. This example disables thinking with a Qwen-specific option; check supported settings before changing models. See the quickstart for dependencies, including Node.js 22 or later for JavaScript.
1 · Set your API key
Open the console and create a key in your workspace.
2 · Set the base URL
Set the Standard Thinking base URL and load your API key from a server-side environment variable.
3 · Pick an enabled model ID
Use GET /v1/models to find the full model IDs available to your key. Check is_ready; the list can include models that are not ready, and readiness does not reserve capacity.
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.STANDARD_API_KEY,
baseURL: "https://api.standardthinking.ai/v1",
});
const response = await client.chat.completions.create({
model: "qwen/qwen3.6-35b-a3b",
messages: [
{ role: "user", content: "Say hello in one short sentence." },
],
max_tokens: 64,
chat_template_kwargs: { enable_thinking: false },
stream: true,
});
for await (const chunk of response) {
process.stdout.write(chunk.choices[0]?.delta?.content || "");
}
Request metadata example.
Illustrative fields for usage and troubleshooting. These sample values are not measured latency or a performance claim.
| Time | Model | Tokens in / out | Latency | Content retention | Cost |
|---|---|---|---|---|---|
| 14:02:11 | deepseek-v4-1-flash |
1,204 / 388 | 412 ms | Not retained | $0.0007 |
| 14:02:09 | glm-5-3 |
2,410 / 1,120 | 1.8 s | Not retained | $0.0083 |
| 14:02:09 | deepseek-v4-1-flash |
640 / 212 | 296 ms | Not retained | $0.0004 |
| 14:02:08 | deepseek-v4-1-flash |
3,980 / 1,502 | 2.4 s | Not retained | $0.0024 |
| 14:02:08 | glm-5-3 |
512 / 96 | 640 ms | Not retained | $0.0011 |
Sample request records. Values are illustrative and do not represent a performance benchmark. Standard API inference is Zero Data Retention; content-free metadata is kept identifiable for up to 90 days from the event.
API data handling.
Request content, operational metadata, and stored features have different data rules.
Request content
Standard API inference, including Standard One, is Zero Data Retention (ZDR): inputs and outputs are not logged. Caching is temporary, and raw text is never written to disk. See the Terms, §5.1.
Operational metadata
Content-free service metadata is identifiable for up to 90 days from the original event; after that, retained data is sufficiently de-identified and is not re-identified. Account, billing, tax, and support records follow separate criteria. See the Terms, §5.2.
Stored features
Features that store or reuse content have separate retention rules and controls.
Model training
We do not use content from ZDR-covered inference to train models, including where a customer separately opts in.
Choose the right access for your work.
Standard API and Product Plan both support products. Coding Plan serves a developer’s own work and, within the plan’s capacity, an application’s users.
Do I need a Product Plan for production?
Standard API also supports products. Consider a Product Plan for monthly base capacity, free extra capacity when available, and optional auto scaling. A data platform for your product’s data is coming to Product Plan but is not released yet.
Can I call every listed model?
The models you can call depend on your workspace and on current capacity. Use GET /v1/models to list model IDs available to your key and inspect is_ready. The list can include models that are not ready; a ready model can still refuse a request if capacity changes.
Is this the same as Coding Plan?
Standard API bills requests by token usage, including requests from coding applications. Coding Plan is a weekly or monthly plan for a developer’s own work and, within the plan’s capacity, an application’s users.