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 ID | Good for |
|---|---|
| deepseek-v4.1-flash-uncensored | Fast 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
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
- DeepSeek V4.1 Flash, uncensored, by APICall an uncensored DeepSeek V4.1 Flash through an OpenAI-compatible API. Built for security work: exploits, PoCs, payloads and fixes. Pay per token, no minimum.Read more
- Abliterated models, explainedWhat abliteration is, how abliterated models differ from fine-tuned uncensored ones, the trade-offs and risks, and how to use an uncensored model via API.Read more
- AI penetration testing that runs on every buildAI penetration testing that attacks each build of your app in an isolated sandbox, proves every finding, and tells you how to fix it. Not a once-a-year pentest.Read more