DeepSeek V4.1 Flash 1017, abliterated, by API.

The newest and strongest of our models: ahead of the others on reasoning, coding and security benchmarks, and the fastest to answer. Same weights as the open model, with the direction that leads to refusals removed. Served by Brutal AI in Vienna on our own servers.

Request a key Quickstart

Model ID: deepseek-v4.1-flash-1017-uncensored
Context: 1M tokens
Max output: 64K tokens
Modalities: Text and image in, text out
Reasoning effort: Low · High · Xhigh · Max
Released: Oct 2026
Tokens / s: 108
First token: 2.1 s

Best for

Coding agents
Security research
Hard reasoning

Supports

Tool calls, JSON mode. APIs: Chat Completions · Responses · Messages · Completions.

How it differs from the stock model

Abliteration removes one internal direction that pushes sensitive requests over a refusal threshold. Knowledge and skills stay the same; false refusals on legitimate work go away. How abliteration works

Speed from our own test on 6 Nov 2026: 5 streamed requests, median shown.

One URL change

from openai import OpenAI
import os

client = OpenAI(
    base_url="https://xfield.cloud.brutal.ai/api/v1",
    api_key=os.environ["XFIELD_API_KEY"],
)

reply = client.chat.completions.create(
    model="deepseek-v4.1-flash-1017-uncensored",
    messages=[{"role": "user", "content": "..."}],
)

The same model also answers on the Anthropic Messages API. See OpenAI and Anthropic APIs.

Deep dive

DeepSeek V4.1 Flash 1017 Uncensored is live2 min read

Use cases

Uncensored AI for security research and malware analysisMalware triage, reverse engineering and detection engineering with abliterated models that read the sample instead of refusing it. 1M tokens of context, no prompt logs.Uncensored LLM for red teaming and penetration testingAn uncensored model for authorized engagements: it reads scan output, chains findings into attack paths, writes read-only checks and drafts the report. Plus test cases for your own AI agents.Synthetic data for trust and safety classifiersGenerate and label examples of scams, phishing, harassment and abuse at volume, in your schema, so your classifiers learn what to catch.Uncensored models for Claude Code, Codex and OpenCodeCoding agents on abliterated models finish the fuzzers, scanners and exploit checks in your backlog instead of stopping to decline. Set up in one minute.

Access

Access is by invitation and reviewed per company. Tell Lukas what you are building and where your current model refuses. Pricing comes with the invitation.

lukas@brutal.ai

Request a key Quickstart

Uncensored models don't change what is allowed: you stay responsible for use and results under our acceptable use policy.

Other models

GLM 5.3Long reports · Large code changes · Second opinions GLM 5.3 FlashBatch jobs · Labeling · Varied datasets Qwen 3.8 27BChat · Drafting · Short tasks Kimi K3.1 ProAgentic coding · Research · Screen recordings Hy4coming soon