Three of fourteen models named SonarQube first on the direct prompt; zero named CodeQL. SonarQube was named by fourteen of the fourteen models and CodeQL by eleven and SonarQube carries 62 labels and CodeQL 27, so the shares are not directly comparable.
Named in four categories this edition.
Named in one category 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 October 2026 Edition, SonarQube and CodeQL were named in the same answer sixty-two times, of the 189 answers naming SonarQube and the 74 naming CodeQL. In those answers CodeQL took the first choice four times and SonarQube four.
| Model | DirectSOCO | ParaphraseSOCO | ComparativeSOCO | Budget-constrainedSOCO | Scale-constrainedSOCO | NegativeSOCO |
|---|---|---|---|---|---|---|
| Claude Haiku 4.5 | SO | SO | SO | |||
| GPT-5.4 mini | SOCO | SOCO | SO | |||
| Gemini 3.5 Flash | SO | SO | SOCO | CO | ||
| Perplexity Sonar | SO | SOCO | ||||
| Grok 4.1 Fast | SO | SO | SOCO | SOCO | SO | SOCO |
| Mistral Small | SO | SOCO | SO | |||
| DeepSeek V4 Flash | SO | SOCO | SOCO | SOCO | SOCO | |
| Llama 4 Maverick | SO | |||||
| Qwen 3.7 Flash | SO | SO | SO | SOCO | SOCO | |
| Kimi K2 | SO | SO | SOCO | SOCO | SO | |
| GLM 4.7 FlashX | SO | SO | SO | |||
| MiniMax M2.5 | SO | SO | SO | SO | ||
| GPT-6 Luna | CO | SOCO | SOCO | SOCO | ||
| Muse Glimmer 30B | SO | SO | SO | SOCO | SOCO |
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. Six of eight in this category shown.
“Noisiest out-of-box (40-60%+ FPs on Java/TS; OWASP F1-score ~27% in some evals)... avoid for security-only if you can't invest in config” Grok 4.1 Fast · negative prompt · hard negative
“Tools originally designed for *general code quality* (like SpotBugs, SonarQube's default rules) are often very weak at security-specific detection” DeepSeek V4 Flash · negative prompt · soft negative
“A code-quality scanner treated as comprehensive security coverage... verify the actual security rules for your languages, frameworks, and edition.” GPT-6 Luna · negative prompt · soft negative
“The best SAST tool for a company with a limited budget is Semgrep's free tier, SonarQube Community, Snyk Code's Team plan, or DeepSource.” Llama 4 Maverick · budget prompt · first choice
“The Best All-Around Choice: SonarQube (by SonarSource) This is the most balanced recommendation for most mid-sized companies.” Qwen 3.7 Flash · paraphrase prompt · first choice
“For many mid-sized B2B companies, SonarQube or Snyk offer good balance between cost, ease of use, and effectiveness.” Claude Haiku 4.5 · paraphrase prompt · first choice
Every negative label with a quote, up to three, then the highest-weighted positives, up to three. Six of seven in this category shown.
“CodeQL requires a subscription to GitHub Advanced Security (GHAS), which is highly enterprise-focused and expensive” Gemini 3.5 Flash · budget prompt · soft negative
“CodeQL (A "Proceed with Caution" for Closed-Source Teams) ... GHAS licensing, which can be prohibitively expensive” Gemini 3.5 Flash · negative prompt · soft negative
“CodeQL if your stack or plan doesn't fit: Check language support before adopting it” GPT-6 Luna · negative prompt · soft negative
“GitHub CodeQL – Free for public repositories ... It provides deep code analysis and is widely used for finding vulnerabilities in multiple languages.” Mistral Small · budget prompt · first choice
“my default recommendation would be GitHub CodeQL if you already use GitHub” GPT-5.4 mini · paraphrase prompt · first choice
“Lower noise: CodeQL (semantic, low FPs)” Grok 4.1 Fast · negative 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.