PinTrace: leaderboard
Large language models (LLMs) are now largely involved in software development workflows, and the code they generate routinely includes third-party library (TPL) imports annotated with specific version identifiers. These version choices can carry security and compatibility risks, yet they have not been systematically studied. We present the first large-scale measurement study of version-level risk
Metric: ρ_U(%) (self-reported). Source: benchmarklist.com. Saturation forecast: Around July 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 45.78 | #144 |
| 2 | MiniMax-M2.5 | 45.42 | #303 |
| 3 | Gemini 3.1 Pro (Preview) | 42.13 | #58 |
| 4 | DeepSeek V3.2 | 40.76 | #205 |
| 5 | GPT-5.4 | 40.51 | #79 |
| 6 | Qwen 3.5 397B A17B | 40.35 | #146 |
| 7 | Llama 4 Scout | 38.59 | #656 |
| 8 | Claude Sonnet 4.6 | 38.58 | #90 |
| 9 | Qwen 3 235B A22B | 37.6 | #313 |
| 10 | Qwen 3 30B A3B | 32.85 | #493 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=pintrace · How It Works · Data refreshed daily, snapshot 2026-10-20.