TraceEval - Python: leaderboard
Metric: F1 (%) on the 769 Python test programs, edge-level F1 of the predicted call graph against execution-verified edges, zero-shot at temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 79.1 |
| 2 | Claude Sonnet 4.6 | 71.3 |
| 3 | GPT-5.4 | 66.6 |
| 4 | DeepSeek V3.2 | 63.1 |
| 5 | GPT-5.4 Mini | 59.7 |
| 6 | Qwen 2.5 Coder 32B Instruct | 57.1 |
| 7 | Llama 3.3 70B Instruct | 53.3 |
| 8 | Gemini 3.1 Pro (Preview) | 49.5 |
| 9 | Qwen 2.5 Coder 7B Instruct | 27.6 |
| 10 | Qwen2.5-Coder-1.5B-Instruct | 16.1 |
Interactive version: theaggregate.ai/benchmark?slug=traceeval-python · How It Works · Data refreshed daily, snapshot 2026-10-07.