DeepSeek 4.1 Flash
deepseek-ai/DeepSeek-V4.1-Flash-1017-abliteratedOur main model for security work.
- Context
- – tokens
- Input
- – / 1M tokens
- Output
- – / 1M tokens
Prices appear here when the model is live in the NEAR AI Cloud catalog.
NEAR AI RED · Cybersecurity
Uncensored AI for cybersecurity. Private by design.
Find the vulnerabilities in your code before attackers do, with AI models that answer security questions instead of refusing them. They run inside secure enclaves on NEAR AI Cloud, so no one sees what you find. Not even NEAR AI.
Recently disclosed
Railway breach leaked customers' secretsPHP flaw reportedly hit a large share of the webCertificate authority key failure enabled fake certificatesNode.js patched a remote code execution flaw
The problem
Attackers already use AI. Your tools refuse to help.
Ask a mainstream model how your code could be hacked, and it refuses. It can't tell a defender from an attacker, so it treats you like one.
Attackers don't have that problem. They run open models with no restrictions.
NEAR AI Cloud fixes that.
The models
Two uncensored models. Open to every account.
Uncensored, also called abliterated, means the model's refusal behaviour is removed. Ask it to find the weak spots in your code, and it does.
nearai · Private TEE
Qwen3.8 27B
Qwen/Qwen3.8-27BFast and low-cost for high-volume scans.
- Context
- 262K tokens
- Input
- $0.04 / 1M tokens
- Output
- $1.35 / 1M tokens
No application. No identity checks. No premium for privacy on NEAR-hosted models.
Why going private
Your vulnerabilities stay yours.
A vulnerability is only safe until someone else knows about it. Run your security research on an AI provider that can read your prompts, and that provider holds a map of how to hack you.
It happens. In October, details of two critical PHP flaws leaked before the planned fix, and the PHP team had to ship its patches early.
security-review.md
Review: payments service, quarter 4
Status: 4 findings open, 1 rated critical
Owner: platform security team
Affected area: internal billing component
Next step: patch, retest, then disclose
Notes: keep this list inside the team until fixed
What NEAR AI can see of your work: nothing readable.
No one can see your prompts.
On NEAR-hosted models, every prompt runs inside a secure enclave built on Intel TDX and NVIDIA confidential computing. The cloud provider, the model provider and NEAR AI cannot see it.
Nothing is logged.
NEAR AI Cloud keeps no logs of prompts or responses.
No silent fallback.
Requests run only inside the enclave. If it's unavailable, they pause instead of going to another provider.
Check it yourself.
Every response is signed inside the enclave. Verify it against Intel and NVIDIA, not against us.
- Response signed inside the enclave…
- Intel TDX quote verified…
- NVIDIA GPU attestation verified (NRAS)…
Not a policy. Not a promise. Proof.
How it works
Start in three steps.
Create a NEAR AI Cloud account.
Pick a model.
DeepSeek 4.1 Flash or Qwen3.8 27B.
Point your tools at the API.
It's OpenAI-compatible, so switching is a base-URL change.
from openai import OpenAI client = OpenAI( base_url=, # point at NEAR AI Cloud api_key=NEAR_AI_CLOUD_API_KEY, ) model = "deepseek-ai/DeepSeek-V4.1-Flash-1017-abliterated"
Why open access
Why we don't gate these models.
These models are already public. Anyone can download them and run them on rented GPUs.
Locking them behind paperwork slows defenders, not attackers. A team under attack can't wait a week for approval.
Proof
Found. Fixed. Kept private.
NEAR AI internal testing
SeverityCritical
StatusFixed
ProjectUndisclosed
Security research only works if the findings stay private until they're fixed. That's why we're interested in what NEAR AI is building: uncensored models we can run privately, with a way to verify that no one else sees what we run, not even the provider.
Grego AI can now use uncensored models without exposing what we find, while safely testing end-to-end proofs of concept in controlled environments without refusals.
NEAR AI Cloud runs in production for Brave, the Government of Bermuda and Venice AI.
NEAR AI RED
For the teams who break things to make them safer.
For red teams, auditors and internal AppSec teams who want to work more closely with NEAR AI on private, uncensored AI. You don't need to join to use the models. Join to get more out of them.
What members get
- RED-0120% off your first month on NEAR AI Cloud.
- RED-02First to try new models as they arrive.
- RED-03Workshops for your security team on AI-assisted security work.
- RED-04Co-marketing. Share what you find, with NEAR AI's channels behind you.
- RED-05Community. Work alongside other security practitioners.
Apply to NEAR AI RED
What members commit to
Use the models only for lawful security research or authorized testing, on systems you own or have written permission to test.
FAQ
Questions security teams ask.
Do I need to join NEAR AI RED to use the models?
No. The models are open to every NEAR AI Cloud account. NEAR AI RED is for teams who want discounts, early access and support.
What does it cost?
Each model's current prices are on its NEAR AI Cloud model page. NEAR AI RED members get 20% off their first month.
Can NEAR AI see my prompts?
No. On NEAR-hosted models, prompts run inside a secure enclave that NEAR AI cannot see into. You can verify this yourself.
What do you keep?
No prompts or responses. Basic account and usage records for security and billing.
What happens if someone misuses the models?
Using the models to cause harm breaks our terms. We investigate suspected abuse and can suspend or close accounts.
Which tools work with it?
Anything that works with the OpenAI API. Change the base URL and keep your workflow.
Find your vulnerabilities before they do.
Uncensored models for security work, running privately on NEAR AI Cloud.