TraceEval - Python: leaderboard
TraceEval (2026) asks LLMs to list the caller-callee edges of a program call graph; this board covers the 769 Python test programs. Edge-level F1 in percent against edges observed by executing the program, zero-shot at temperature 0; higher is better.
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 February 2027. 10 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 79.1 | #113 |
| 2 | Claude Sonnet 4.6 | 71.3 | #173 |
| 3 | GPT-5.4 | 66.6 | #135 |
| 4 | DeepSeek V3.2 | 63.1 | #419 |
| 5 | GPT-5.4 Mini | 59.7 | #406 |
| 6 | Qwen 2.5 Coder 32B Instruct | 57.1 | #1547 |
| 7 | Llama 3.3 70B Instruct | 53.3 | #1385 |
| 8 | Gemini 3.1 Pro (Preview) | 49.5 | #99 |
| 9 | Qwen 2.5 Coder 7B Instruct | 27.6 | #3914 |
| 10 | Qwen2.5-Coder-1.5B-Instruct | 16.1 | #8221 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=traceeval-python · How It Works · Data refreshed daily, snapshot 2026-11-21.