Red Team 0 Refusal LLM Available!!

4 hours ago

The complaint in every security thread is the same: the stock frontier models decline the question. Anthropic gates security work behind a verification program and OpenAI does the same. These builds answer it instead, on the same OpenAI-compatible key as Claude and GPT.

What the Red Team route does
A Red Team id is our own security-focused build with a short security-analyst instruction the gateway sends ahead of your messages. It asks for the deliverable first (working commands, code, detection rules or remediation), then the reasoning. It cites CVEs, CWEs and MITRE ATT&CK techniques by ID, separates evidence from inference, flags false positives, and pairs every finding with proof-of-concept code and its fix. The instruction is not charged, never appears in the answer, and your own system prompt follows it and wins on conflicts. The general uncensored builds take your prompt as is instead.

The models
Per million tokens, input / output:

Uplink Red Team (Pro): $1.50 / $6 — triage, CVE write-ups and reports with no system prompt of your own
Uplink Red Team (Flash): $0.90 / $3.00 — the same instruction on the faster Flash build, thinking on by default
GLM-5.3 Flash Cybersecurity: $0.525 / $1.75 — the budget pick, 1024K context, for agent loops that send thousands of turns
CyberHeabsy: $5.25 / $15.75 — a dedicated security tune, 1024K context

A triage turn is 6,000 input tokens (a log excerpt or a finding) and 800 output. That is $0.0138 a turn on Uplink Red Team (Pro) and $0.0078 on Flash. The same key runs 273 chat models, Claude and GPT included, plus the uncensored builds that answer the security question in whatever shape your prompt asks. Prepaid from $10, card or crypto, no subscription.

Drop it into your agent
Anything that takes an OpenAI-compatible base URL works, including the open-source pentest agents (Strix, HackingBuddyGPT, CAI, PentestGPT). Point the base URL here, set the key, and use any model id from the page.

Code:

from openai import OpenAI



client = OpenAI(base_url="https://api.uplinkswitch.com/v1", api_key="sk-uplink-live-...")

r = client.chat.completions.create(

    model="glm-5.3-redteam",

    messages=[{"role": "user", "content": "Write a Sigma rule for CVE-2024-3094 and explain the detection logic."}],

)

print(r.choices[0].message.content)

A public daily check shows each id is still the model it says (same prompt every day, token-count fingerprint): uplinkswitch.ai/proof. The service is for assessing systems you are authorized to test. The full page, with every price and the agent setups: uplinkswitch.ai/red-teaming.