TraceEval - Java: leaderboard

TraceEval (2026) asks LLMs to list the caller-callee edges of a program call graph; this board covers the 711 Java 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 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 February 2027. 10 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.674.9#113
2Claude Sonnet 4.672.2#173
3DeepSeek V3.269.3#419
4GPT-5.459.5#135
5GPT-5.4 Mini54.1#406
6Llama 3.3 70B Instruct33.9#1385
7Qwen 2.5 Coder 32B Instruct32.5#1547
8Qwen 2.5 Coder 7B Instruct23.5#3914
9Gemini 3.1 Pro (Preview)13.7#99
10Qwen2.5-Coder-1.5B-Instruct5.1#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-java · How It Works · Data refreshed daily, snapshot 2026-11-21.