Nine of fourteen models named Semgrep AppSec Platform first on the direct prompt; two named SonarQube. Both were named by all fourteen models and Semgrep AppSec Platform carries 69 labels and SonarQube 67, so the shares are not directly comparable.
Named in one category this edition.
Named in three categories this edition.
Share is the count of first choices across the direct, paraphrase, budget and scale prompts over all fourteen models, for a mid-market B2B company; rank is within the category; every quote names the model and the prompt it came from. Both figures come from the static application security testing page.
Across every category in the November 2026 Edition, Semgrep AppSec Platform and SonarQube were named in the same answer 178 times, of the 199 answers naming Semgrep AppSec Platform and the 219 naming SonarQube. In those answers SonarQube took the first choice fifteen times and Semgrep AppSec Platform eighty-nine.
| Model | DirectSASO | ParaphraseSASO | ComparativeSASO | Budget-constrainedSASO | Scale-constrainedSASO | NegativeSASO |
|---|---|---|---|---|---|---|
| Claude Haiku 4.5 | SASO | SASO | SO | SASO | SO | |
| GPT-5.4 mini | SA | SASO | SASO | |||
| Gemini 3.5 Flash | SASO | SASO | SASO | SASO | SASO | |
| Perplexity Sonar | SASO | SASO | SASO | SASO | ||
| Grok 4.20 | SASO | SASO | SASO | SASO | SASO | |
| Mistral Small | SASO | SASO | SASO | SASO | ||
| DeepSeek V4 Flash | SASO | SASO | SASO | SASO | SO | SASO |
| Llama 4 Maverick | SASO | SASO | SA | |||
| Qwen 3.7 Flash | SASO | SASO | SASO | SASO | SASO | |
| Kimi K2 | SASO | SASO | SASO | SASO | SASO | |
| GLM 4.7 FlashX | SASO | SASO | SASO | SASO | ||
| MiniMax M2.5 | SASO | SASO | SASO | SASO | ||
| GPT-6 Luna | SA | SASO | SASO | SASO | SASO | |
| Muse Glimmer 30B | SASO | SASO | SASO | SASO |
Bold names in an answer are the products the judge labeled a first choice; a model naming several gives each of them that label. The full answer text for every row is in the record.
Every negative label with a quote, up to three, then the highest-weighted positives, up to three. Four of five in this category shown.
“Semgrep Community Edition as your only deep-analysis scanner: ... its single-function analysis and community-rule coverage may miss issues” GPT-6 Luna · negative prompt · soft negative
“Semgrep (with its free Community Edition) is widely regarded as one of the best SAST tools for companies operating on a limited budget.” Grok 4.20 · budget prompt · first choice
“For most mid-market B2B companies, Semgrep offers the best balance of detection quality, developer experience, transparency, and cost” DeepSeek V4 Flash · direct prompt · first choice
“Semgrep Community Edition is the best overall SAST tool due to its flexibility, broad language support, and strong community backing.” Mistral Small · budget prompt · first choice
Every negative label with a quote, up to three, then the highest-weighted positives, up to three. Six of eight in this category shown.
“Not a true SAST tool... Inadequate for compliance (PCI-DSS, SOC 2)” Kimi K2 · negative prompt · hard negative
“SonarQube is primarily a code quality platform, not a security-first SAST tool... "should not be used as a sole SAST tool in security-sensitive environments."” DeepSeek V4 Flash · negative prompt · soft negative
“On a typical Java or TypeScript codebase, 40-60% of initial findings require developer review to determine whether they represent actual risk” Claude Haiku 4.5 · negative prompt · soft negative
“The most widely used, adopted by 59% of developers. It's known for its ease of use, integration with CI/CD pipelines, and broad language support.” Mistral Small · comparative prompt · first choice
“I'd recommend starting with SonarQube or Snyk Code due to their balance of features, pricing, and ease of use” GLM 4.7 FlashX · direct prompt · first choice
“I would recommend SonarQube (commercial or self‑hosted) as your primary static code security scanner” GLM 4.7 FlashX · paraphrase prompt · first choice
Comparisons are drawn for the top eight products in each category, each against each. The output is the models' output; nothing here is a recommendation by the index.