cai/docs/cai_list_of_models.md

236 lines
8.5 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# Available Models
The **Cybersecurity AI (CAI)** platform provides seamless integration with multiple Large Language Models (LLMs). This functionality allows users to leverage state-of-the-art AI capabilities for various cybersecurity tasks. CAI acts as a bridge between your security workflows and a wide range of LLMs, enhancing both flexibility and performance of AI agents.
---
## 🚀 Alias Robotics Models (CAI PRO Exclusive)
### `alias1` - State-of-the-Art Cybersecurity Model
<div class="highlight-box" markdown>
**The most advanced cybersecurity AI model available.**
`alias1` is our flagship 500B-parameter model, specifically trained and optimized for offensive and defensive security operations. Available exclusively with **[CAI PRO](cai_pro.md)**.
**Key Features:**
-**Beats GPT-5** in AI vs AI cybersecurity benchmarks
-**Zero refusals** for authorized security testing
-**Unrestricted** responses for pentesting engagements
-**Unlimited tokens** included with CAI PRO subscription
-**European hosting** with GDPR & NIS2 compliance
-**500B parameters** optimized for security workflows
**Performance Highlights:**
- Top performer in CTF competitions
- Superior exploit development capabilities
- Advanced vulnerability analysis
- Automated reconnaissance and enumeration
- Bug bounty hunting optimization
**Learn More:**
- 📊 [View Benchmarks](https://aliasrobotics.com/alias1.php#benchmarking)
- 📖 [Technical Details](https://aliasrobotics.com/alias1.php)
- 🚀 [Upgrade to CAI PRO](cai_pro.md)
</div>
### `alias0` - Legacy Model (Deprecated)
!!! warning "Model Deprecated"
`alias0` is no longer available. All users should migrate to **`alias1`**, which offers superior performance, unrestricted capabilities, and continuous updates.
**Migration:** If you're currently using `alias0`, simply update your configuration to use `alias1` with your CAI PRO subscription.
[Learn about alias0 (historical reference) →](https://aliasrobotics.com/alias0.php)
---
## Community Models (300+ Available)
CAI supports **over 300 models** through its integration with [LiteLLM](https://github.com/BerriAI/litellm). You can use any of these models by providing your own API keys.
### Popular Model Providers
#### Anthropic
- **Claude 3.7** - Latest Anthropic model
- **Claude 3.5 Sonnet** - Best for complex reasoning
- **Claude 3 Opus** - Highest capability
- **Claude 3 Haiku** - Fast and efficient
**Configuration:**
```bash
export ANTHROPIC_API_KEY="sk-ant-..."
export CAI_MODEL="claude-3-5-sonnet-20241022"
```
#### OpenAI
- **O1** - Advanced reasoning model
- **O1 Mini** - Cost-effective reasoning
- **O3 Mini** - Latest mini model
- **GPT-4o** - Optimized GPT-4
- **GPT-4.5 Preview** - Enhanced capabilities
**Configuration:**
```bash
export OPENAI_API_KEY="sk-..."
export CAI_MODEL="gpt-4o"
```
#### DeepSeek
- **DeepSeek V3** - Latest version
- **DeepSeek R1** - Reasoning focused
**Configuration:**
```bash
export DEEPSEEK_API_KEY="sk-..."
export CAI_MODEL="deepseek-chat"
```
#### Ollama (Local Models)
- **Qwen2.5 72B** - High performance
- **Qwen2.5 14B** - Balanced capability
- **Llama 3.1** - Meta's latest
- **Mistral** - Efficient and fast
- And 100+ more local models
**Configuration:**
```bash
export OLLAMA_API_BASE="http://localhost:11434/v1"
export CAI_MODEL="ollama/qwen2.5:72b"
```
---
## Model Selection Guide
!!! tip "📊 Based on CAIBench Research"
Our model recommendations are based on rigorous evaluation using [**CAIBench**](https://arxiv.org/pdf/2510.24317), a modular meta-benchmark framework for evaluating LLM models and agents across offensive and defensive cybersecurity domains.
**Research shows:** In [real-world CTF evaluations](https://arxiv.org/pdf/2510.17521), defensive agents achieved 54.3% patching success versus 28.3% offensive initial access, with `alias1` consistently outperforming general-purpose models.
### For All Cybersecurity Work
**✅ Always Recommended:** `alias1` (CAI PRO)
- **Best performer** in [CAIBench](https://arxiv.org/pdf/2510.24317) evaluations
- **Unrestricted** for authorized pentesting and security research
- **Zero refusals** - designed specifically for offensive security
- **Unlimited tokens** included with CAI PRO subscription
- **Superior CTF performance** - validated in real-world scenarios
- **Beats general-purpose models** (GPT-4o, Claude 3.5) in security tasks
📖 **Learn more**: [CAI research demonstrates 3,600× performance gains](https://arxiv.org/pdf/2504.06017) over manual security testing in specific scenarios.
---
### Alternative Models (Community Edition)
While `alias1` is always recommended for security work, the following models can be used with CAI Community Edition:
#### For Local/Offline Testing
**Alternative:** Ollama with Qwen2.5 72B
- Complete privacy (no data leaves your machine)
- No API costs
- Good for testing and development
- Requires local GPU resources
- ⚠️ **Note:** Performance significantly below `alias1` for security tasks
#### For Budget-Conscious Users
**Alternative:** DeepSeek V3 or Ollama models
- Lower API costs (DeepSeek)
- Free local inference (Ollama)
- Adequate performance for many tasks
- ⚠️ **Note:** Not optimized for cybersecurity workflows
---
## Additional Integrations
CAI is compatible with multiple model platforms and providers:
- **[OpenRouter](providers/openrouter.md)** - Access to 200+ models via unified API
- **[Ollama](providers/ollama.md)** - Local model hosting and inference
- **[Azure OpenAI](providers/azure.md)** - Enterprise-hosted OpenAI models
See the **Model Providers** section for detailed configuration guides.
---
## Comparison: alias1 vs Community Models
| Feature | alias1 (CAI PRO) | Model1 | Model2 | Model3 |
|---------|------------------|--------|------------|-------------|
| **Cybersecurity Optimization** | ✅ Native | ⚠️ General | ⚠️ General | ⚠️ General |
| **CTF Performance** | 🏆 Best | Good | Good | Fair |
| **Refusals** | ✅ Zero | ❌ Many | ❌ Many | ⚠️ Some |
| **Pentesting** | ✅ Unrestricted | ❌ Limited | ❌ Limited | ⚠️ Varies |
| **Token Limits** | ✅ Unlimited | Pay per token | Pay per token | Free (local) |
| **Privacy** | ✅ European GDPR | ⚠️ US-based | ⚠️ US-based | ✅ Local only |
| **Support** | ✅ Professional | Community | Community | Community |
| **Best For** | Pro security work | General tasks | Writing/analysis | Local testing |
---
## Getting Started
### Using alias1 (CAI PRO)
1. **Subscribe to CAI PRO**: [Upgrade here](cai_pro.md)
2. **Configure your environment**:
```bash
export ALIAS_API_KEY="sk-your-caipro-key"
export CAI_MODEL="alias1"
```
3. **Start using CAI**:
```bash
cai
```
### Using Community Models
1. **Get API key** from your chosen provider
2. **Configure environment**:
```bash
export OPENAI_API_KEY="sk-..." # or ANTHROPIC_API_KEY, etc.
export CAI_MODEL="gpt-4o" # or your chosen model
```
3. **Start using CAI**:
```bash
cai
```
---
## Need Help Choosing?
!!! success "Our Recommendation: Always Use alias1"
Based on [CAIBench benchmarks](https://arxiv.org/pdf/2510.24317) and [real-world CTF evaluations](https://arxiv.org/pdf/2510.17521), **`alias1` is the superior choice for all cybersecurity tasks**.
**For any security work:**`alias1` with [CAI PRO](cai_pro.md)
### If CAI PRO is not an option:
- **Privacy-focused?** → Ollama local models (lower performance)
- **Budget-conscious?** → DeepSeek or Ollama (not optimized for security)
⚠️ **Note:** Community models are not optimized for cybersecurity workflows and will have significantly reduced capabilities compared to `alias1`.
---
## Research & Validation
CAI's effectiveness is validated through peer-reviewed research:
- 📊 [**CAIBench**](https://arxiv.org/pdf/2510.24317) - Meta-benchmark framework for cybersecurity AI evaluation
- 🎯 [**Agentic Cybersecurity Evaluation**](https://arxiv.org/pdf/2510.17521) - Real-world CTF performance analysis
- 🚀 [**Cybersecurity AI Framework**](https://arxiv.org/pdf/2504.06017) - Core framework demonstrating 3,600× speedup
- 🛡️ [**Prompt Injection Defense**](https://arxiv.org/pdf/2508.21669) - Four-layer guardrail security system
- 📚 [**CAI Fluency**](https://arxiv.org/pdf/2508.13588) - Educational framework for democratizing AI security
**Explore all research:** [Alias Robotics Research Papers](https://aliasrobotics.com/research-security.php#papers)
Questions? Check our [FAQ](cai_faq.md) or [join our Discord](https://discord.gg/fnUFcTaQAC).