Towards Evaluation of Implicit Software World Models in Coding LLMs: leaderboard
This SWE-bench Verified–based evaluation probes coding LLMs’ implicit software world models by predicting execution-related signals such as peak memory, wall-clock time, and ranked profiler outputs at method and line granularity. The evaluated model uses text input. The pinned metric is F1. Source: BenchmarkList’s attributed result table.
Metric: F1 (%). Source: benchmarklist.com. Saturation forecast: Around February 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 84.2 | #33 |
| 2 | GPT-OSS-120B | 60.4 | #346 |
| 3 | Claude Sonnet 4.6 | 54.5 | #94 |
| 4 | GPT-5 Mini | 49.7 | #193 |
| 5 | Claude Opus 4.7 | 48.6 | #51 |
| 6 | Qwen 3.5 397B A17B | 45.1 | #154 |
| 7 | GPT-5.2 | 39.3 | #115 |
| 8 | GPT-5.4 | 37.8 | #83 |
| 9 | Claude Haiku 4.5 | 28.2 | #287 |
| 10 | Qwen 3 235B A22B | 14.5 | #323 |
| 11 | Qwen 3 30B A3B | 4.9 | #496 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=towards-evaluation-of-implicit-software-world-models-in-coding-llms · How It Works · Data refreshed daily, snapshot 2026-11-05.