Free Kimi K3 API: A Free Reasoning Model for Chat, Code, and Roleplay

Kimi K3 reasons through a problem before it writes its answer. That makes it strong at code, math, planning, and long, consistent roleplay, and slower to start than a non-reasoning model. Its model page lists its current context and output limits: models.

1. Sign up at freetheai.org/signup and confirm your email.
2. Create a key under Dashboard > API keys and copy all of it.
3. Check in at freetheai.org/checkin. Free models stay unlocked until 00:00 UTC.
4. Point your app at https://api.freetheai.org/v1 (or https://api.freetheai.org/v1/chat/completions for apps that want the full address) and pick fta/kimi/k3.

from openai import OpenAI

client = OpenAI(base_url="https://api.freetheai.org/v1", api_key="ftai_your_key_here")
reply = client.chat.completions.create(
    model="fta/kimi/k3",
    messages=[{"role": "user", "content": "Plan a three-step study schedule for Go."}],
)
print(reply.choices[0].message.content)

Thinking is on by default. For quick chat, turn it off per request:

{
  "model": "fta/kimi/k3",
  "reasoning_effort": "none",
  "messages": [{ "role": "user", "content": "Hi!" }]
}

With thinking on, keep streaming turned on in your app. Chat apps such as JanitorAI give up on a reply that sends nothing for a while and show "No response from bot"; streaming sends the words as they come. See fixing "No response from bot".

• Temperature and Top P: K3 accepts only fixed values, so FreeTheAI sets them for you. Moving those sliders in your app changes nothing, and does not cause errors.
• Max tokens: raise it for long replies, or send 0 to get the model's maximum.
• System prompts: supported. Character cards and long system prompts work as usual.
• Tool calls: supported, for coding agents and function calling.