PerfCodeBench: leaderboard
Large language models (LLMs) can often generate functionally correct code, but their ability to produce efficient implementations for performance-critical systems tasks remains limited. Existing code benchmarks mainly emphasize correctness or algorithmic problem solving, while realistic systems-level optimization is still underexplored. To address this gap, we introduce PerfCodeBench, an executabl
Metric: CGRE (self-reported). Source: benchmarklist.com. Saturation forecast: Around September 2027. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 73.99 | #98 |
| 2 | Qwen 3.6 Max | 66.32 | #96 |
| 3 | GPT-5.4 | 66.27 | #79 |
| 4 | Claude Opus 4.5 | 65.09 | #83 |
| 5 | Qwen 3.6 27B | 65.07 | #170 |
| 6 | DeepSeek V4 Pro | 64.3 | #102 |
| 7 | Kimi K2.6 | 62.14 | #104 |
| 8 | Claude Sonnet 4.5 | 61.6 | #142 |
| 9 | Gemini 3.1 Pro (Preview) | 61.46 | #58 |
| 10 | DeepSeek V4 Flash | 61.25 | #137 |
| 11 | Seed 2.0 Lite | 60.19 | #145 |
| 12 | Qwen 3.6 Plus | 56.46 | #120 |
| 13 | Kimi K2 | 44.73 | #243 |
| 14 | Qwen 3.6 35B A3B | 44.38 | #226 |
| 15 | Llama 4 Maverick | 43.72 | #464 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=perfcodebench · How It Works · Data refreshed daily, snapshot 2026-10-20.