CodePanel

PHP code review: which models catch the bugs

Of the 17 models ranked on PHP code, 5 models tied for the most known bugs: 100%. 9 of the other 12 models have 95% intervals that overlap theirs. Gemini 3.5 Flash-Lite caught the fewest, 0%. Together they caught 64% of known bugs in PHP code, against 60% for the same 17 across the whole benchmark.

Updated 2026-11-04Full leaderboardMethodology

Results

Known defects caught, with 95% intervals. The tick is the model's recall over every bug.

PHP code review: which models catch the bugs
ModelCaught hereEvery bug
Claude Opus 5.5100%90%
GPT-6.1 Sol100%89%
Gemini 4 Argon100%82%
GPT-5.6 Sol100%72%
Claude Haiku 5.5100%64%
DeepSeek V4 Flash (Workers AI)78%[33%–100%]49%
DeepSeek V4 Pro (Workers AI)69%[33%–100%]54%
Claude Sonnet 5.567%[0%–100%]77%
GPT-6 Luna67%[0%–100%]67%
Gemini 3.1 Pro67%[0%–100%]67%
Gemini 3.8 Flash50%[0%–100%]68%
GLM-5.3 (Workers AI)44%[0%–100%]61%
Kimi K2.7 Code (Workers AI)42%[25%–75%]38%
GLM-5.3 Flash (Workers AI)33%[0%–100%]61%
GPT-OSS 120B (Workers AI)33%[0%–67%]40%
Gemma 4 26B A4B (Workers AI)33%[0%–100%]26%
Gemini 3.5 Flash-Lite0%11%

Against the rest of the benchmark

PHP was the sixth-easiest of the 12 languages for these models.

Claude Haiku 5.5 gains the most here: 100%, against 64% over every bug.

GLM-5.3 Flash drops the most: 33%, against 61% over every bug.

A test PR that split the models

The change “Fix collection filtering API for IN/NOT IN comparisons that require type conversions” in doctrine/orm, a change of 100 to 300 lines that landed in October 2025, introduced a logic bug that a later upstream commit fixed.

Claude Opus 5.5, GPT-6.1 Sol, Gemini 4 Argon, GPT-5.6 Sol, Gemini 3.8 Flash, GPT-6 Luna, Claude Haiku 5.5, Gemini 3.1 Pro and DeepSeek V4 Flash (Workers AI) caught it; GLM-5.3, DeepSeek V4 Pro (Workers AI), GPT-OSS 120B (Workers AI) and Kimi K2.7 Code (Workers AI) caught it in part; Claude Sonnet 5.5, GLM-5.3 Flash, Gemma 4 26B A4B and Gemini 3.5 Flash-Lite missed it.