Multi-LCB - Pass@5 T0.6 - Python: leaderboard
Evaluates code generation across 12 programming languages. Appendix Table 8 reports pass@5 at temperature 0.6, distinct from the headline pass@1 experiment. Reasoning mode and effort identifiers are retained. This series reports Multi-LCB - Pass@5 T0.6 - Python using Pass@5 (%). Source: the original paper result table.
Metric: Pass@5 (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-OSS-120B (Medium) | 84.7 | #346 (GPT-OSS-120B) |
| 2 | GPT-OSS-20B (Medium) | 79.2 | #523 (GPT-OSS-20B) |
| 3 | Qwen 3 235B A22B (Thinking) | 71.1 | #323 (Qwen 3 235B A22B) |
| 4 | GPT-OSS-120B (Low) | 70.5 | #346 (GPT-OSS-120B) |
| 5 | Qwen 3 32B (Thinking) | 70.1 | #449 (Qwen 3 32B) |
| 6 | Qwen 3 14B (Reasoning) | 68.4 | #542 (Qwen 3 14B) |
| 7 | Qwen 3 30B A3B (Thinking) | 64.4 | #496 (Qwen 3 30B A3B) |
| 8 | Qwen 3 8B (Thinking) | 61 | #678 (Qwen 3 8B) |
| 9 | GPT-OSS-20B (Low) | 59.9 | #523 (GPT-OSS-20B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=multi-lcb-pass-5-t0-6-python · How It Works · Data refreshed daily, snapshot 2026-11-05.