An uncensored LLM API built for security work

General-purpose models stop at the useful part: "I can't help write an exploit." Swarm Computers serves an uncensored DeepSeek V4.1 Flash, so your tools and agents can finish the job on systems you are authorised to test.

Why security teams need an uncensored model

Offensive security is full of requests that look harmful out of context: write a payload, weaponise this CVE, bypass this filter. Mainstream models refuse many of them, refuse inconsistently, or water down the answer. That breaks automated pipelines, where a single refusal can stall an agent mid-test.

An uncensored or abliterated model answers the request as asked. For a red team, a bug bounty hunter or an AppSec engineer validating a fix, that is the difference between a tool that works and one that needs babysitting.

The model

Model IDGood for
deepseek-v4.1-flash-uncensoredFast exploit drafting, payload generation, PoCs and patches, agent loops

DeepSeek V4.1 Flash is a fast open-weight model, served here with its refusals removed. The standard version is on the same API as deepseek-v4.1-flash. More models will be added as they pass our testing.

Drop-in for the tools you already use

request.sh
curl https://swarmcomputers.com/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4.1-flash-uncensored",
    "messages": [{"role": "user",
      "content": "Write a PoC for the IDOR in this handler, then the fix."}]
  }'

The API follows the OpenAI chat completions format, so existing SDKs, agent frameworks and tools that accept a custom base URL work by changing two settings. Billing is per token with no minimum.

Use cases

  • Pentest and red-team agents that must not stall on a refusal.
  • Turning a scanner finding into a working proof of concept, then a patch.
  • Writing detection rules (YARA, Sigma) from real attack code.
  • Bug bounty triage and report writing.
  • Generating adversarial test data for your own AI guardrails.

The API is for testing systems you own or are authorised to assess.

Questions

What does uncensored mean here?
The model's trained-in refusal behaviour has been removed, so it answers security requests directly. It does not mean anything goes: our terms limit use to authorised testing and research.
Is an uncensored model less capable?
Removing refusals can cost a little quality on some tasks. DeepSeek V4.1 Flash is strong enough that the trade is worth it for security work.
How much does it cost?
Per million tokens: $0.12 (provisional) input, $0.015 (provisional) cached input and $0.525 (provisional) output. There is no minimum spend, and you pay from prepaid credits.
Is it compatible with OpenAI SDKs?
Yes. Set the base URL and API key and keep the rest of your code.

Related

Create a key and make your first call.