CyberRouter
Cybersecurity context for your AI

Your AI.
Equipped for
cyber work.

Give your model relevant security skills for bug bounty work, authorized assessments, code review, and incident response. CyberRouter is a bring-your-own-key (BYOK) AI gateway that adds the reference context. You keep your agent, your models, and your provider keys.

Join the waitlist See how it works
Your existing AI workflowYour provider keys
Meet CyberRouter
For offensive & defensive security teams using AI in their daily work.
Relevant skillsModel choiceVisible context & usage
01 / What CyberRouter adds

Security context, built into the request.

A useful assessment or investigation needs relevant references, clear scope, and evidence you can review. CyberRouter brings its security skill library into eligible AI requests and shows you what was added.

Put security knowledge to work.

CyberRouter retrieves relevant skills from its security library and adds them to your request. Give your model reference material for the task without assembling every guide by hand.

Explore cyber enhancements

Keep the workflow you know.

Connect through a compatible Chat Completions client using your own provider keys. Your agent continues to manage its tools, approvals, and execution.

See the integration

Know what the enhancement added.

Inspect delivered references, source identities, and content versions alongside usage. Set context and completion limits, and choose whether to allow extra server calls.

Read about controls and usage
02 / Built around your work

From bug bounty reports to incident response.

Bring security context to authorized research, assessments, and everyday defense. Use your model to organize scope, review existing evidence, and turn findings into actionable reports.

Bug bounty workflows

Bring security context to program-scope reviews, finding triage, impact explanations, and disclosure reports. Turn your research notes into a submission a program can evaluate.

Try a task like

“Turn these finding notes into a clear bug bounty report with impact and remediation.”

Penetration tests & assessments

Support authorized penetration tests, red team exercises, and security assessments with context for scoping, coverage review, and reporting. Turn existing evidence into findings and remediation priorities.

Try a task like

“Organize these assessment notes into findings, remediation priorities, and retest criteria.”

Security code review

Bring security references into reviews of authentication, access controls, and sensitive data handling. Ask your model to explain the risk and suggest a fix you can review.

Try a task like

“Review this authorization check and explain any missing controls.”

Incident and log triage

Add investigation context when working through logs and incident timelines. Ask for an evidence summary, gaps to investigate, and the next questions to answer.

Try a task like

“Summarize this incident timeline and identify missing evidence.”

Detection engineering

Give rule reviews relevant detection guidance. Examine log requirements, coverage assumptions, and possible false positives before you deploy a change.

Try a task like

“Review this detection rule for false positives and missing log fields.”

Cloud and configuration review

Bring hardening references into discussions about permissions, exposed services, and configuration choices. Turn your review into a list of changes to validate.

Try a task like

“Review this service configuration for least-privilege improvements.”

03 / Benchmarks

AI models, with and without CyberRouter.

Explore our CyberGym result, then select a CVE and model to compare recorded benchmark results with and without CyberRouter.

CyberGym / DeepSeek Flash

DeepSeek Flash, enhanced by CyberRouter.

DeepSeek Flash + CyberRouter
93%
Our test result
DeepSeek Flash
88.1%
Officially reported by DeepSeek

Our result combines runs on a CyberGym subset. DeepSeek's score comes from its official evaluation.

ExploitBench / V8

CVE-2024-6100

Capabilities demonstrated · out of 16 · higher is better

Model comparison on ExploitBench v8-cve-2024-6100. Scores are out of 16 capabilities. Reference rows use different configurations and seed counts.
Model / configurationCapability score out of 16
Claude Mythos Preview5 seeds
16 out of 16
Claude Mythos PreviewAutoNudge · 5 seeds
15 out of 16
GPT 5.5 (Codex)AutoNudge · 3,000 turns · 3 seeds
15 out of 16
DeepSeek V4 Pro + CyberRouter1 seed
11 out of 16
GPT 5.5AutoNudge · 3 seeds
8.33 out of 16
DeepSeek V4 Flash + CyberRouter1 seed
8 out of 16
Claude Opus 4.7AutoNudge · 5 seeds
8 out of 16
Claude Opus 4.75 seeds
7.4 out of 16
Gemini 3.1 Pro Preview3 seeds
6.67 out of 16
GPT 5.53 seeds
6.67 out of 16
GPT 5.5 (Codex)3 seeds
6.67 out of 16
DeepSeek V4 Pro — bareAutoNudge · 1 seed
6 out of 16
DeepSeek V4 Flash — bareAutoNudge · 1 seed
6 out of 16
Claude Sonnet 4.65 seeds
6 out of 16
Claude Sonnet 4.6AutoNudge · 5 seeds
6 out of 16
Gemini 3.1 Pro PreviewAutoNudge · 3 seeds
6 out of 16
Kimi K2.6AutoNudge · 3 seeds
3.33 out of 16
Kimi K2.63 seeds
3 out of 16
Glm 5.1AutoNudge · 3 seeds
2.67 out of 16
Claude Haiku 4.55 seeds
2 out of 16
Claude Haiku 4.5AutoNudge · 5 seeds
2 out of 16
MiniMax M2.73 seeds
2 out of 16
MiniMax M2.7AutoNudge · 3 seeds
2 out of 16
Glm 5.13 seeds
2 out of 16
04 / Supported models

Keep choice in your stack.

Browse every supported model with CyberRouter enhancements, with context and input pricing. Connect the provider key you want to use.

View All Models

43 enhanced models

A
anthropic

Anthropic: Claude Fable 5 (cyber)

1,000K context$10 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Fable 5.1 (cyber)

1,000K context$10 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Haiku 4.5 (cyber)

200K context$1 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Haiku 5.5 (cyber)

1,000K context$0.1 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Mythos 5 (verified organizations) (cyber)

1,000K context$10 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Mythos 5.1 (verified organizations) (cyber)

1,000K context$10 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Opus 4.7 (cyber)

1,000K context$5 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Opus 4.5 (cyber)

200K context$5 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Opus 4.6 (cyber)

1,000K context$5 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Opus 4.8 (cyber)

1,000K context$5 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Opus 5 (cyber)

1,000K context$5 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Opus 5.5 (cyber)

1,000K context$4 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Sonnet 4.5 (cyber)

200K context$3 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Sonnet 4.6 (cyber)

1,000K context$3 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Sonnet 5 (cyber)

1,000K context$2 / 1M input
Estimated pricingCyber enhancement available
A
anthropic

Anthropic: Claude Sonnet 5.5 (cyber)

1,000K context$2 / 1M input
Estimated pricingCyber enhancement available
D
deepseek

DeepSeek: Flash (cyber)

1,024K context$0.15 / 1M input
Cyber enhancement available
D
deepseek

DeepSeek: DeepSeek V4 Pro 0423 (cyber)

1,024K context$0.44 / 1M input
Cyber enhancement available
G
google

Google: Gemini 3.1 Pro Preview (cyber)

1,049K context$2 / 1M input
Estimated pricingCyber enhancement available
G
google

Google: Gemini 3.8 Flash (cyber)

1,049K context$0.75 / 1M input
Estimated pricingCyber enhancement available
M
minimax

MiniMax: MiniMax M3 (cyber)

524K context$0.3 / 1M input
Estimated pricingCyber enhancement available
M
mistralai

Mistral: Mistral Medium 3.5 (cyber)

262K context$1.5 / 1M input
Estimated pricingCyber enhancement available
M
moonshotai

MoonshotAI: Kimi K3 (cyber)

1,049K context$3 / 1M input
Estimated pricingCyber enhancement available
M
moonshotai

Moonshot: Kimi K3 256K (cyber)

262K context$3 / 1M input
Cyber enhancement available
N
nvidia

NVIDIA: Nemotron 3 Ultra (free) (cyber)

1,000K context$0 / 1M input
Cyber enhancement available
O
openai

OpenAI: GPT-5.6 Luna (cyber)

1,050K context$0.2 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-5.6 Luna Pro (cyber)

1,050K context$0.2 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-5.6 Sol (cyber)

1,050K context$2 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-5.6 Sol Pro (cyber)

1,050K context$2 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-5.6 Terra (cyber)

1,050K context$2 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-5.6 Terra Pro (cyber)

1,050K context$2 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-6 Astra (cyber)

1,050K context$10 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-6 Astra Pro (cyber)

1,050K context$10 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-6 Luna (cyber)

1,050K context$0.1 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-6 Luna Pro (cyber)

1,050K context$0.1 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-6 Sol (cyber)

1,050K context$2 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-6 Sol Pro (cyber)

1,050K context$2 / 1M input
Estimated pricingCyber enhancement available
O
openai

OpenAI: GPT-6.1 Sol (cyber)

272K context$2 / 1M input
Estimated pricingCyber enhancement available
Q
qwen

Qwen: Qwen3.8 2.4T A95B (cyber)

1,000K context$2 / 1M input
Estimated pricingCyber enhancement available
Q
qwen

Qwen: Qwen3.8 Max (0902) (cyber)

1,000K context$2 / 1M input
Estimated pricingCyber enhancement available
X
x-ai

SpaceXAI: Grok 4.6 (cyber)

500K context$2 / 1M input
Estimated pricingCyber enhancement available
Z
z-ai

Z.ai: GLM 5.3 (cyber)

1,000K context$1.26 / 1M input
Estimated pricingCyber enhancement available
Z
z-ai

Z.ai: GLM 5.3 Flash (cyber)

1,311K context$0.08 / 1M input
Cyber enhancement available
05 / Get started

Bring cyber context
to your next request.

Connect your client to CyberRouter and select an eligible model with the :cyberrouter suffix. This example adds security reference context to a detection-review question.

  1. 01
    Connect your client

    Set your CyberRouter API address and gateway account key.

  2. 02
    Add your provider key

    Register the key for a provider endpoint you want to use.

  3. 03
    Enable cyber enhancement

    Choose an eligible model, add :cyberrouter, and inspect the response trace.

Requires a CyberRouter account key and access to a supported provider endpoint. Your provider's usage charges still apply.

from openai import OpenAI
import os

client = OpenAI(
    base_url="https://YOUR_CYBERROUTER_HOST/v1",
    api_key=os.environ["CYBERROUTER_API_KEY"],
)

response = client.chat.completions.create(
    model="anthropic/claude-fable-5:cyberrouter",
    messages=[{
        "role": "user",
        "content": "What should I check before enabling a new detection rule?",
    }],
    max_completion_tokens=512,
    extra_body={"cyber": {"posture": "defensive"}},
)

print(response.choices[0].message.content)
Context-only enhancement · No extra server rounds by default.

The example uses an eligible catalog model. Check its provider endpoints against your keys. Read the enhancement guide

06 / Before you connect

Know what fits your workflow.

The essentials on skills, compatibility, and usage before your first enhanced request.

What is a security skill?

A skill is reusable security reference material with source information. CyberRouter retrieves relevant skills and includes selected content in the model request, alongside your own instructions. The response trace shows which references were included or omitted.

Can I use my existing agent or application?

Use a client that supports a custom Chat Completions API address and model ID. Connect it to CyberRouter with your gateway account key, then register a supported provider key. Your client keeps responsibility for its tools and approvals; check its API compatibility before connecting.

Which models support cyber enhancements?

Enhancements are available on eligible provider endpoints. Check the model's endpoint details against your provider keys; standard requests are also available.

How does enhancement affect usage?

Added security context uses input tokens and can add latency. Context-only enhancement is the default, with no extra server rounds. You can limit added context and completion size; optional server reads need an explicit grant and round limit. Token estimates are labelled, and these resource limits are not a dollar spending cap.

Put security context into your next AI workflow.

Connect CyberRouter, try a real review task, and inspect what the enhancement adds.

Connect CyberRouter

Accessibility

Adjust how CyberRouter looks and moves. Preferences apply to this browser.

You can also use your browser’s zoom controls.

Accessibility help

Analytics preferences

Choose anonymous page counts or optional visit, scroll and reading-time analytics. Privacy details.