DeepSeek V4.1 Flash, uncensored

Two ways of removing the refusals

Both start from the same DeepSeek model and differ in how it was opened up.

• DeepSeek V4.1 Flash 1017 is abliterated by hand: the refusal direction is found in the model's activations and taken out of the weights. A million tokens of context.
• DeepSeek Flash Lite was made with Heretic, a tool that does the same job automatically, searching for the lightest edit that stops refusals while changing the model's other answers as little as it can. Half a million tokens of context, and the lowest cached price on this site.

Both read images, call tools and reason before they answer.

Why people use it for long chats

Cached input on Flash Lite is $0.03 per million tokens. A chat resends its history every turn, so in a long roleplay or a coding session most of what you send is cached, and the running cost stays close to the price of the new words only. A 2,000 token prompt with a 600 token reply costs about $0.0016 before any caching.

Reasoning effort

DeepSeek V4.1 Flash thinks before it replies. Send reasoning_effort in the request to trade speed for care. low is the right setting for chat, where a quick reply matters more than a long think.

Calling it

curl https://uncensoredchats.com/v1/chat/completions \
  -H "Authorization: Bearer $UNCENSORED_AI_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek-v4.1-flash-lite-1017-xploded",
       "messages": [{"role": "user", "content": "Hello"}]}'

Keep reading

• Kimi K3.1 Pro, uncensored→
• GLM 5.3, uncensored→
• Qwen3.8 27B, uncensored→
• The uncensored AI API→
• Set up SillyTavern→
• Set up opencode→
• Set up Claude Code→
• Abliterated vs uncensored→
• Cost calculator→
• Uncensored AI roleplay→

Updated 12 November 2026. Prices are read from the live price list.

Start chattingAll models and prices