TraceEval: leaderboard
Evaluating whether large language models (LLMs) can recover execution-relevant program structure, rather than only produce code that passes tests, remains an open problem. Existing code benchmarks emphasize test-passing outputs, from standalone programming tasks (HumanEval, MBPP, LiveCodeBench) to repository repair (SWE-Bench); this is useful, but offers limited diagnostic signal about which progr
Metric: Average F1 (self-reported). Source: benchmarklist.com. Saturation forecast: Around May 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 72.9 | #63 |
| 2 | Claude Sonnet 4.6 | 64.5 | #90 |
| 3 | DeepSeek V3.2 | 62.1 | #205 |
| 4 | GPT-5.4 | 60 | #79 |
| 5 | GPT-5.4 Mini | 56.5 | #212 |
| 6 | Qwen 2.5 Coder 32B Instruct | 46.6 | #528 |
| 7 | Llama 3.3 70B Instruct | 44.3 | #534 |
| 8 | Gemini 3.1 Pro (Preview) | 31.8 | #58 |
| 9 | Qwen 2.5 Coder 7B Instruct | 27.1 | #951 |
| 10 | Qwen2.5-Coder-1.5B-Instruct | 9.5 | #1460 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=traceeval · How It Works · Data refreshed daily, snapshot 2026-10-20.