Stock model

Kimi K3 Fast

The vendor's original weights, served unmodified on shared GPUs.

Stock models are the vendor's original weights, not abliterated: they keep the vendor's own refusals. Our safety check reads text only, so images and video sent to them are not screened. Prompts and outputs, images and video included, are never stored (privacy policy).

Specs

Model id: kimi-k3-fast
Context length: 1,048,576 tokens
Max output: 1,048,576 tokens (4,096 on trial accounts)
Input: Text and images. Audio, video and file inputs are rejected. Our safety check reads text only, so images are not screened.
Runs on: The vendor's original weights, on shared GPUs operated by our infrastructure provider under the same zero-data-retention configuration.
Reasoning: Per request: reasoning_effort on Chat Completions, thinking on Messages, reasoning.effort on Responses.
Endpoints: /v1/chat/completions, /v1/completions, /v1/messages, /v1/responses (stateless)

Price

Input: $4.05
Cached input: $0.405
Output: $20.25

USD per 1M tokens, billed from prepaid credit.

Uptime, last 30 days

Status page

100%

26 of 26 canary checks passed (one every 5 minutes).

Last canary: passed, time to first token 649 ms, total 700 ms.

Speed, last 24 h

—

Speed is measured, not guaranteed: the median output tokens per second after the first token, and the median time to first token (TTFT), over streamed requests in the last 24 h. "—" until a model has 20 of them.

Quickstart

Create a key in the dashboard, then call kimi-k3-fast from any OpenAI or Anthropic client.

curl https://api.abliteratedllm.dev/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k3-fast",
    "stream": true,
    "messages": [{"role": "user", "content": "Write the opening line of a noir detective story."}]
  }'