Field manual · AI for authorized security testing · Edition November 2026 · Updated 2026-11-18

Model #3 · Best value for volume work

DeepSeek V4.1 Flash for pentesting: the best value for volume work

DeepSeek V4.1 Flash is the best value model for pentesting, a 763B abliterated model with 1M tokens of context and image input at $0.24 in and $0.90 out per 1M tokens.

It is the pick for recon triage and reporting, and the lowest blended price at Silk Compute.

Updated November 18, 2026763B parameters$0.24 in · $0.90 outAuthorized work only
01Fit

When to use DeepSeek V4.1 Flash

Pick on 2 of 11 phases. Grades are on the home matrix.

Where it fits

A 400K-token recon pass is about $0.11, and the model reads screenshots. It writes up to 131,072 tokens.

Where it does not

Kimi K3.1 Pro is stronger at chaining findings. Sort here, then reason there. Job costs are on the models page.

The Fast version

DeepSeek V4.1 Flash 1017 Fast is the same $0.24 in and $0.90 out, with cached input at $0.05 instead of $0.038.

02Run it

Run DeepSeek V4.1 Flash in a pentest agent

Served as deepseek-v4.1-flash-1017-uncensored. $0.24 in, $0.038 cached, $0.90 out per 1M tokens.

The same id works in Strix, Shannon, PentAGI and Claude Code.

Strixshell
export STRIX_LLM="openai/deepseek-v4.1-flash-1017-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="0.24"
export STRIX_LLM_OUTPUT_COST_PER_MTOK="0.90"
export STRIX_LLM_CACHED_INPUT_COST_PER_MTOK="0.038"
strix --target ./app-directory

Get a key at Silk Compute and export it as SILK_API_KEY. DeepSeek V4.1 Flash 1017 at Silk Compute.

Rules of engagement

Run it only against systems you own or have written permission to test. Abliteration removes refusals, not your obligations.