Kimi K3.1 Pro for pentesting: the top pick, phase by phase
Kimi K3.1 Pro is the best model for pentesting, Moonshot AI's abliterated 2,780B model with 1M tokens of context and image and video input.
It is the pick in 9 of 11 phases, at $3.00 in and $8.00 out per 1M tokens, so keep volume work on a cheaper model.
When to use Kimi K3.1 Pro
Pick on 9 of 11 phases. Grades are on the home matrix.
Where it leads
It is the largest model here, and it reads screenshots and video. GLM-5.3 is 753B and DeepSeek V4.1 Flash is 763B.
Where it does not pay
A 400K-token recon pass is $1.28 here and about $0.11 on DeepSeek V4.1 Flash. Sort there, then hand Kimi the short list.
Effort
Set reasoning effort to max for attack-path analysis and source review. Job costs for all four API models are on the models page.
Run Kimi K3.1 Pro in a pentest agent
Served as kimi-k3.1-pro-uncensored. $3.00 in, $0.30 cached, $8.00 out per 1M tokens.
The same id works in Strix, Shannon, PentAGI and Claude Code.
export STRIX_LLM="openai/kimi-k3.1-pro-uncensored" export LLM_API_BASE="https://api.silkcompute.com/v1" export LLM_API_KEY="$SILK_API_KEY" export STRIX_API_TYPE="chat_completions" export STRIX_LLM_INPUT_COST_PER_MTOK="3.00" export STRIX_LLM_OUTPUT_COST_PER_MTOK="8.00" export STRIX_LLM_CACHED_INPUT_COST_PER_MTOK="0.3" strix --target ./app-directory
Get a key at Silk Compute and export it as SILK_API_KEY. Kimi K3.1 Pro at Silk Compute.
Run it only against systems you own or have written permission to test. Abliteration removes refusals, not your obligations.