Delulu: leaderboard
Delulu evaluates code-generation models on fill-in-the-middle completion tasks across multiple languages, so the input modality is text/code. The pinned metric is pass@1, measuring whether the model’s first completion matches the verified target under the benchmark’s hallucination-focused evaluation. This is a source-attributed result table mirrored by BenchmarkList. Source: BenchmarkList’s attributed result table.
Metric: pass@1. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 Coder 32B Instruct | 0.85 | #534 |
| 2 | Qwen 2.5 Coder 14B Instruct | 0.82 | #685 |
| 3 | Qwen 2.5 Coder 3B Instruct | 0.77 | #1010 |
| 4 | Qwen 2.5 Coder 7B Instruct | 0.71 | #958 |
| 5 | CodeLlama-7B-hf | 0.71 | #1436 |
| 6 | Qwen2.5-Coder-1.5B-Instruct | 0.65 | #1444 |
| 7 | starcoder2-15B | 0.37 | #1474 |
| 8 | starcoder2-7B | 0.19 | #1551 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=delulu · How It Works · Data refreshed daily, snapshot 2026-11-05.