TypeScript code review: which models catch the bugs
Of the 17 models ranked on TypeScript code, Claude Opus 5.5 caught the most known bugs: 100%. 9 of the other 16 models have 95% intervals that overlap its own. Gemma 4 26B A4B and Gemini 3.5 Flash-Lite tied for the fewest, 33%.
| Model | Caught here | Every bug | Difference |
|---|---|---|---|
| Claude Opus 5.5 | 100% | 90% | +10 pts |
| GPT-6.1 Sol | 83%[50%–100%] | 89% | −6 pts |
| Gemini 4 Argon | 83%[50%–100%] | 82% | +1 pt |
| Claude Sonnet 5.5 | 83%[50%–100%] | 77% | +6 pts |
| Gemini 3.8 Flash | 83%[50%–100%] | 68% | +15 pts |
| GPT-6 Luna | 83%[50%–100%] | 67% | +16 pts |
| Claude Haiku 5.5 | 83%[50%–100%] | 64% | +19 pts |
| DeepSeek V4 Pro (Workers AI) | 79%[46%–100%] | 54% | +25 pts |
| GLM-5.3 (Workers AI) | 78%[53%–96%] | 61% | +17 pts |
| Kimi K2.7 Code (Workers AI) | 75%[53%–94%] | 38% | +37 pts |
| GPT-5.6 Sol | 67%[33%–100%] | 72% | −5 pts |
| Gemini 3.1 Pro | 67%[33%–100%] | 67% | 0 pts |
| GLM-5.3 Flash (Workers AI) | 57%[22%–90%] | 61% | −4 pts |
| DeepSeek V4 Flash (Workers AI) | 46%[13%–83%] | 49% | −3 pts |
| GPT-OSS 120B (Workers AI) | 42%[14%–72%] | 40% | +2 pts |
| Gemma 4 26B A4B (Workers AI) | 33%[8%–67%] | 26% | +7 pts |
| Gemini 3.5 Flash-Lite | 33%[0%–67%] | 11% | +22 pts |
On average these models caught 69% of known bugs in TypeScript code, more often than the 60% the same 17 caught across the whole benchmark: the fourth-easiest of the 12 languages.
Kimi K2.7 Code (Workers AI) gains the most here: 75%, against 38% over every bug.
GPT-6.1 Sol drops the most: 83%, against 89% over every bug.
The change “feat(v4): parse input containing reference cycles” in colinhacks/zod, a change of 100 to 300 lines that landed in August 2026, introduced a resource-handling bug that a later upstream commit fixed.
Claude Opus 5.5, GPT-6.1 Sol, Claude Sonnet 5.5, GPT-5.6 Sol, GPT-6 Luna and Claude Haiku 5.5 caught it; GLM-5.3 and Kimi K2.7 Code (Workers AI) caught it in part; Gemini 4 Argon, Gemini 3.8 Flash, Gemini 3.1 Pro, GLM-5.3 Flash, DeepSeek V4 Pro (Workers AI), DeepSeek V4 Flash (Workers AI), GPT-OSS 120B (Workers AI), Gemma 4 26B A4B and Gemini 3.5 Flash-Lite missed it.
Every model that reviewed “Properly handle redirects in prerendered pages” (remix-run/react-router) caught its bug.