SuperCLUE-Code3 (May 2024) - Overall: leaderboard
SuperCLUE-Code3 evaluates Chinese code generation with functional unit tests, as a Chinese extension of HumanEval. May 2024 edition (2024年5月): difficulty-weighted score (难度等级加权得分) over basic, intermediate and advanced problems, the column this edition ranks by. Each dated workbook is a separate evaluation round, kept apart from the current board. Score 0-100, higher is better.
Metric: Difficulty-Weighted Score. Source: www.superclueai.com. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 71.68 | #357 |
| 2 | GPT-4 Preview (0125) | 68 | #500 |
| 3 | GPT-4 | 63.74 | #459 |
| 4 | Llama 3 70B Instruct | 62.56 | #640 |
| 5 | DeepSeek V2 | 62.52 | #549 |
| 6 | GPT-3.5 Turbo (0125) | 55.51 | #975 |
| 7 | deepseek-coder-6.7B-instruct | 47.77 | #1316 |
| 8 | Gemini 1.0 Pro | 46.5 | #900 |
| 9 | Qwen-14B-Chat | 24.67 | #1161 |
| 10 | Llama 2 13B Chat | 6.06 | #1414 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=superclue-code3-may-2024-overall · How It Works · Data refreshed daily, snapshot 2026-11-05.