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

Model #2 · Strong second for code and text

GLM-5.3 for pentesting: the strong second for code and text

GLM-5.3 is the second-best model for pentesting, Z.ai's abliterated 753B model at $2.00 in and $4.00 out per 1M tokens.

It reads text only, so it fits source review, cloud policy and tooling better than a test full of screenshots.

Updated November 18, 2026753B parameters$2.00 in · $4.00 outAuthorized work only
01Fit

When to use GLM-5.3

Graded against the other four models.

Where it fits

Source, infrastructure code and IAM policy are text. GLM-5.3 is strong there and not the pick. Strix's quick start uses it as the example model. (Strix README)

Where it does not

It cannot read an image. Send that step to Kimi K3.1 Pro or DeepSeek V4.1 Flash. Job costs are on the models page.

02Run it

Run GLM-5.3 in a pentest agent

Served as glm-5.3-uncensored. $2.00 in, $0.30 cached, $4.00 out per 1M tokens.

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

Strixshell
export STRIX_LLM="openai/glm-5.3-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="2.00"
export STRIX_LLM_OUTPUT_COST_PER_MTOK="4.00"
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. GLM-5.3 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.