AI Models

CyberStrike supports 9,000+ models across 230+ providers through a live catalog powered by models.dev. New models appear automatically — no upgrade needed.

Run /models in the TUI to browse what’s available with your configured providers, or use --model provider/model-id at launch.

How Model Selection Works

CyberStrike uses two model slots:

Set them in cyberstrike.json:

{
  "model": "anthropic/claude-sonnet-5-5",
  "small_model": "anthropic/claude-haiku-5-5"
}

Or at launch:

cyberstrike --model anthropic/claude-opus-5-5

Or switch mid-session with /models in the TUI — it shows all models from your configured providers, sorted by capability.

Model Capabilities

Every model in the catalog carries metadata that CyberStrike uses for routing and feature gating:

Reasoning Options

Models with reasoning support may offer different ways to control it:

Configure reasoning in cyberstrike.json:

{
  "provider": {
    "anthropic": {
      "options": {
        "thinking": {
          "type": "enabled",
          "budgetTokens": 10000
        }
      }
    }
  }
}

Anthropic (Claude)

All Claude models support tool calling, reasoning, and file attachments. Claude is recommended for CyberStrike’s security analysis workflows due to its strong reasoning and instruction-following capabilities.

All Claude models accept text and image input. Reasoning options vary: Opus and Sonnet models support effort levels and budget tokens; Haiku supports toggle and budget tokens.

Anthropic Subscription (Claude Pro/Max)

If you have a Claude Pro or Max subscription, you can authenticate via OAuth instead of an API key:

1. Run /connect in the TUI
2. Select Anthropic and choose Login with Anthropic account
3. Authorize in the browser

This uses your subscription quota — no separate API billing.

OpenAI (GPT)

Google (Gemini)

Google’s Gemini models offer the largest context windows in the industry — up to 1M tokens — with multimodal input support including text, images, audio, video, and PDF.

Gemma open models (gemma-4-31b-it, gemma-4-26b-a4b-it) are also available via Google at zero cost.

xAI (Grok)

Grok models offer large output limits (up to 500K tokens) and competitive pricing.

Mistral

Groq

Groq provides ultra-fast inference on specialized hardware. Most models are free or very low cost.

DeepInfra

DeepInfra hosts a wide range of open-source and commercial models at competitive prices. 76+ models available.

Highlights:

Together AI

Together AI runs 32+ models with competitive pricing and fast inference.

Cohere

Cerebras

Cerebras provides ultra-fast inference. Currently hosts 2 models:

Alibaba (Qwen)

Alibaba’s Qwen family includes 59+ models with strong multilingual support. Set DASHSCOPE_API_KEY to use.

Highlights:

Amazon Bedrock

Bedrock is a multi-model gateway — it gives you access to Claude, Nova, GPT, Grok, Mistral, DeepSeek, Qwen, and more through your AWS account. 199 models available.

CyberStrike handles cross-region inference prefixes automatically. See the AI Providers page for authentication setup.

Key models available through Bedrock include:

• All Anthropic Claude models (Opus, Sonnet, Haiku, Fable)
• Amazon Nova family (Micro, Lite, Pro, Nova 2 Lite)
• OpenAI GPT-5.x and GPT-6.x series
• xAI Grok models
• Mistral models
• DeepSeek V3.2
• Qwen models

OpenRouter

OpenRouter aggregates models from many providers under a single API key. Set OPENROUTER_API_KEY and access models from Anthropic, OpenAI, Google, Meta, and dozens more.

Model IDs use the OpenRouter format: openrouter/provider/model-name.

GitHub Copilot

Access models through your existing Copilot subscription — no separate API key needed.

CyberStrike automatically routes Claude models through the Anthropic Messages API instead of the OpenAI-compatible endpoint — this happens transparently.

See the AI Providers page for authentication setup.

Google Vertex AI

Access Google’s models and Anthropic Claude through your GCP project. Requires GOOGLE_CLOUD_PROJECT and Application Default Credentials.

Two provider IDs:

• google-vertex — Gemini and other Google models
• google-vertex-anthropic — Claude models via Vertex

Continental multi-regions (eu, us) use Regional Endpoint Platform domains automatically.

GitLab Duo

Access AI models through your GitLab instance. Set GITLAB_TOKEN or authenticate via OAuth.

Additional Hosted Providers

Custom & Self-Hosted Models

Any OpenAI-compatible endpoint works with CyberStrike. Define it in cyberstrike.json:

{
  "provider": {
    "my-llm": {
      "name": "My Local LLM",
      "api": "http://localhost:8000/v1",
      "models": {
        "my-model": {
          "name": "My Model",
          "limit": { "context": 128000, "output": 16384 },
          "tool_call": true
        }
      }
    }
  },
  "model": "my-llm/my-model"
}

Common self-hosted setups:

Filtering Models

Control which models appear with whitelist and blacklist:

{
  "provider": {
    "anthropic": {
      "whitelist": ["claude-opus-5-5", "claude-sonnet-5-5", "claude-haiku-5-5"],
      "blacklist": ["claude-fable-5", "claude-fable-5-1"]
    }
  }
}

Or disable entire providers:

{
  "disabled_providers": ["openrouter", "groq"],
  "enabled_providers": ["anthropic", "openai"]
}

Live Catalog

The model catalog refreshes automatically from models.dev every hour. You can also force a refresh by restarting CyberStrike.

To override the catalog source:

CYBERSTRIKE_MODELS_URL=https://my-mirror.example.com cyberstrike
CYBERSTRIKE_MODELS_PATH=/path/to/local/models.json cyberstrike
CYBERSTRIKE_DISABLE_MODELS_FETCH=true cyberstrike