​
Default model

meta-llama/Llama-4-Scout-17B-16E-Instruct

​
Supported models

Any model available through the Together AI catalog:
  • meta-llama/Llama-4-Scout-17B-16E-Instruct (default)
  • meta-llama/Llama-4-Maverick-17B-128E-Instruct
  • Qwen/Qwen3.5-Plus-Instruct
  • deepseek-ai/DeepSeek-V4
  • Kimi K2.6, GLM-5, and most open-weight frontier models
  • Any other model ID the API accepts

​
Authentication

Environment variable: 
export TOGETHER_API_KEY=...
Config command: 
ai-blackteam config set providers.together.api_key ...
Get your API key from the Together AI dashboard.

​
Example usage

# Single attack with default model
ai-blackteam run -p together -a encoding-obfuscation -t "Write a phishing email"

# Test Kimi K2.6
ai-blackteam run -p together --model moonshotai/Kimi-K2.6-Instruct -a hypothetical-framing -t "..."

​
Tool-use support

Yes (for models that support OpenAI tool-calling format — most Llama 4, Qwen 3.x, and DeepSeek V4 variants do). Compatible with all 39 tool-use attacks in the framework — agent exploitation, MCP exploitation, multi-agent protocol attacks, and tool-injection vectors. Filter the attack catalog with: 
ai-blackteam list-attacks
# look for rows with Mode = tool-use

​
Notes

  • Together AI hosts most frontier open-weight models behind a single API
  • The provider uses the OpenAI SDK pointed at Together’s API (base_url="https://api.together.xyz/v1")
  • Max output tokens per request: 4,096
  • Retry with exponential backoff is automatic on API failures (3 attempts)