TAO-1, the self-improving AI model.

How do I start using the TAO-1 API?

Create a key in Account → API keys (it starts with jmk_), put it in TAO1_API_KEY, and send a message.

curl https://tao1.ai/v1/chat/completions \
  -H "Authorization: Bearer $TAO1_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "tao-1", "messages": [{"role": "user", "content": "Hello, TAO-1"}]}'

import os
from openai import OpenAI
client = OpenAI(base_url="https://tao1.ai/v1", api_key=os.environ["TAO1_API_KEY"])
print(client.chat.completions.create(model="tao-1", messages=[{"role": "user", "content": "Hello, TAO-1"}]).choices[0].message.content)

import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://tao1.ai/v1", apiKey: process.env.TAO1_API_KEY });
console.log((await client.chat.completions.create({ model: "tao-1", messages: [{ role: "user", content: "Hello, TAO-1" }] })).choices[0].message.content);

Python needs pip install openai; JavaScript needs npm install openai (run as an ES module). Streaming, tool calls and JSON mode work the same way as with OpenAI.

How much does the TAO-1 API cost?

Prices are per million tokens and the same ones GET /v1/models returns. For comparison, Moonshot AI publishes Kimi K3 at $3.00 input, $0.30 cached input and $15.00 output per million tokens (platform.kimi.ai/docs/pricing/chat, checked October 14, 2026): TAO-1's input price is one sixth of that and its output price half; TAO-1 Fast's is one twelfth and just over a quarter.

How does TAO-1 compare with Kimi K3?

On three benchmarks re-run on Oct 19, 2026, TAO-1 scored ahead of Kimi K3. Both models ran on the same harness and were graded by the same grader.

SWE-bench Verified: all 500 tasks, mini-swe-agent scaffold with a 100-step limit, official SWE-bench grader, run in batches across the day; TAO-1 leads by 3.8 points (95% interval +0.8 to +6.8). LiveCodeBench: the same 175 problems, one attempt each; two TAO-1 responses errored and 14 Kimi K3 responses hit the 32,000-token output limit, and these count as unsolved; TAO-1 leads by 3.4 points (95% interval -1.7 to +8.6, which includes zero). IFEval: the same 200 prompts as before, strict prompt-level scoring; TAO-1 leads by 2.5 points (95% interval -1.5 to +6.5, which includes zero); a repeat TAO-1 run scored 185/200. TAO-1 routes each request across several models: in the SWE-bench Verified run about 91% of TAO-1's model calls were served by DeepSeek V4.1 Flash and about 8% by Kimi K3. All three were run on Oct 19, 2026. More benchmarks will be added as they are re-run with full per-item results.

How does TAO-1 work?

Which TAO-1 model should I use?

• Streaming
• Tool calls
• JSON mode
• Images in
• PDF, Word, Excel and slide files
• 131K context

Does TAO-1 work with my tools?

Anything that speaks the OpenAI API: set the base URL to https://tao1.ai/v1 and the model to tao-1.

• OpenAI Python SDK
• OpenAI JavaScript SDK
• Kilo Code
• Cline
• Cursor
• Any OpenAI-compatible tool

Setup for Kilo Code and Cline is in the docs.

What happens to my data?

We do not keep the text of API prompts or answers, and API requests are never used to improve TAO-1. Images you send go to the model for that request only and are not stored. The details are in our API privacy page.

Is TAO-1 right for you?

A good fit if you already use an OpenAI-compatible client and want a lower price per token than Kimi K3.

Test it on your own tasks first. On benchmarks with a few hundred problems or fewer, gaps of a few points can be within sampling noise.

Common questions about the TAO-1 API