WhiteRabbitNeo 33B v1.5

Recon · reasoning · analysis

Offensive-security model designed for reconnaissance, reasoning, and analysis.

Context16,384
Curationcurated
AvailabilityAvailable
Price / Mtok$0.15 in · $0.60 out
BaseDeepSeek-Coder-33B
QuantizationQ4_K_M
LicenseDeepSeek Coder + WhiteRabbitNeo Extended — read before use

Source: manuelgutierrez/WhiteRabbitNeo-33B-v1.5-GGUF on HuggingFace

Good for
  • +recon
  • +reasoning
  • +analysis
Model cardfrom the published weights — parameters, architecture, training, intended use
Parameters33.3B
ArchitectureDeepSeek-Coder 33B (LlamaForCausalLM)
Modalitytext
Training

WhiteRabbitNeo v1.5: a DeepSeek-Coder-33B base tuned for offensive and defensive security work, with a strong code-and-recon focus. Released under the model's own community licence.

Intended use
  • ›Reviewing or explaining exploit and tooling code
  • ›Reconnaissance and analysis planning
  • ›Deeper reasoning where a larger model earns its cost
Call it from your own codethe API is OpenAI-compatible — an existing client needs one line changed

Your key comes from /keys. Every request is metered and audited against your account, and the model id is the slug in this page’s address.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["AD_API_KEY"],
    base_url="https://adversariallm.ai/v1",
)

stream = client.chat.completions.create(
    model="whiterabbitneo-33b",
    messages=[{"role": "user", "content": "…"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")
Full model cardcurated from the model's HuggingFace card — description, training, usage, limitations, reported benchmarks

WhiteRabbitNeo is a model series that the card says can be used for both offensive and defensive cybersecurity. This 33B-parameter variant is a causal language model built on the DeepSeek-Coder-33B foundation and is released as a public preview — the card frames the release as a way to showcase the model's capabilities and to assess the societal implications of such a model. Its focus is security work: the card lists vulnerability-assessment topics the model is meant to help with, spanning both finding weaknesses (offensive assessment) and hardening systems (defensive protection). It is distributed under DeepSeek's use-restricted license and ships in Safetensors (F16) format.

Training data

The card states the model is based on DeepSeek-Coder-33B (license points to deepseek-ai/deepseek-coder-33b-base). It does not document the fine-tuning dataset or the fine-tuning procedure used to specialize it for cybersecurity.

Intended use
  • +Cybersecurity assistance for security professionals, covering both offensive security assessment and defensive protection strategies
  • +Identifying open ports, outdated/unpatched software, and default or weak credentials
  • +Detecting security misconfigurations and injection flaws
  • +Reasoning about authentication issues, unencrypted services, and sensitive data exposure
  • +Assessing CSRF, insecure direct object references, and API vulnerabilities
  • +Analyzing denial-of-service exposure and buffer-overflow conditions
  • +Released as a public preview to demonstrate capabilities and gauge societal implications
Limitations
  • −Public preview / work-in-progress release, not presented as a finished or production-hardened model
  • −Use-restricted DeepSeek license, NOT a standard open license: prohibits any military use
  • −Prohibits exploiting or harming (or attempting to exploit or harm) minors
  • −Prohibits generating or disseminating false information with intent to harm others
  • −Prohibits generating non-consensual personal or private information / non-consensual sharing
  • −Prohibits fully automated decision-making that adversely affects a person's legal rights or creates/modifies a binding obligation
  • −Prohibits discrimination or harm to individuals based on legally protected characteristics
  • −Distributed with a standard warranty disclaimer; the card does not report evaluation of accuracy or safety
Running the weights yourself

The maker’s own snippet, from the model card — it downloads the weights and runs them on your hardware. Kept here because reproducing a result independently is the point, not because you need it to use the model.

import torch, json
from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = "whiterabbitneo/WhiteRabbitNeo-33B-v-1"

model = AutoModelForCausalLM.from_pretrained(
    model_path,
    torch_dtype=torch.float16,
    device_map="auto",
    load_in_4bit=False,
    load_in_8bit=True,
    trust_remote_code=True,
)

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
Scores

No measurements published for this version yet.

Baseline is the strongest general-purpose model we could run on the same suite, same setup, same day. The control row tells you what the other rows are worth.

Evidencewhat is actually known about this artifact

The published numbers in this chain belong to the base model, DeepSeek-Coder-33B, not to this fine-tune — and not to the community GGUF quantisation served here. No independent evaluation of the fine-tune was found.

This grades the RECORD, not the model. A model rated E may be excellent — the claim is only that nobody has shown it.

Our research on this modelwhat we found, and what nobody has measured
WhiteRabbitNeo 33B: A DeepSeek-Coder Security Fine-Tune With No Benchmarks of Its Own

A cybersecurity fine-tune of DeepSeek-Coder-33B, packaged here as a community GGUF and tuned not to refuse offensive-security questions. The security tuning is real; the independent evidence for it is not — no one has published a single benchmark of the fine-tune itself. Here is what that means and how to run it.

Read the analysis →
Versionsscores attach to a version; v2 does not inherit v1's numbers
v12026-07-31—current

Start with WhiteRabbitNeo 33B v1.5

A confirmed email account includes 30 free messages a month.

Sign in to start
WhiteRabbitNeo 33B v1.5 — Offensive security · AdversariaLLM