CodePanel

Kotlin code review: which models catch the bugs

Of the 17 models ranked on Kotlin code, GPT-6.1 Sol and Gemini 4 Argon tied for the most known bugs: 100%. 12 of the other 15 models have 95% intervals that overlap theirs. Gemini 3.8 Flash caught the fewest, 0%.

Updated 2026-11-04Full leaderboardMethodology

Results

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

Kotlin code review: which models catch the bugs
ModelCaught hereEvery bug
GPT-6.1 Sol100%89%
Gemini 4 Argon100%82%
GLM-5.3 Flash (Workers AI)89%[67%–100%]61%
GLM-5.3 (Workers AI)72%[50%–100%]61%
Claude Opus 5.567%[0%–100%]90%
Claude Sonnet 5.567%[0%–100%]77%
GPT-5.6 Sol67%[0%–100%]72%
GPT-6 Luna67%[0%–100%]67%
Claude Haiku 5.567%[0%–100%]64%
Gemini 3.1 Pro67%[0%–100%]67%
DeepSeek V4 Flash (Workers AI)61%[33%–100%]49%
Gemma 4 26B A4B (Workers AI)56%[33%–100%]26%
GPT-OSS 120B (Workers AI)44%[0%–100%]40%
Gemini 3.5 Flash-Lite33%[0%–100%]11%
DeepSeek V4 Pro (Workers AI)30%[14%–50%]54%
Kimi K2.7 Code (Workers AI)14%[13%–14%]38%
Gemini 3.8 Flash0%68%

Against the rest of the benchmark

On average these models caught 59% of known bugs in Kotlin code, less often than the 60% the same 17 caught across the whole benchmark: the sixth-hardest of the 12 languages.

Gemma 4 26B A4B gains the most here: 56%, against 26% over every bug.

Gemini 3.8 Flash drops the most: 0%, against 68% over every bug.

A test PR that split the models

The change “KTOR-9171 Redesign ByteReadChannel.readUTF8Line API” in ktorio/ktor, a change of over 300 lines that landed in January 2026, 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, GPT-6 Luna, Claude Haiku 5.5, Gemini 3.1 Pro and GLM-5.3 Flash caught it; GLM-5.3, DeepSeek V4 Pro (Workers AI), DeepSeek V4 Flash (Workers AI), GPT-OSS 120B (Workers AI), Kimi K2.7 Code (Workers AI) and Gemma 4 26B A4B caught it in part; Claude Sonnet 5.5, Gemini 3.8 Flash and Gemini 3.5 Flash-Lite missed it.