TraceEval - Java: leaderboard
Metric: F1 (%) on the 711 Java 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 | 74.9 |
| 2 | Claude Sonnet 4.6 | 72.2 |
| 3 | DeepSeek V3.2 | 69.3 |
| 4 | GPT-5.4 | 59.5 |
| 5 | GPT-5.4 Mini | 54.1 |
| 6 | Llama 3.3 70B Instruct | 33.9 |
| 7 | Qwen 2.5 Coder 32B Instruct | 32.5 |
| 8 | Qwen 2.5 Coder 7B Instruct | 23.5 |
| 9 | Gemini 3.1 Pro (Preview) | 13.7 |
| 10 | Qwen2.5-Coder-1.5B-Instruct | 5.1 |
Interactive version: theaggregate.ai/benchmark?slug=traceeval-java · How It Works · Data refreshed daily, snapshot 2026-10-07.