NL2Repo: leaderboard
NL2Repo evaluates long-horizon coding capabilities including repository-level understanding, where models must generate or modify code across entire repositories from natural language specifications.
Metric: Score (self-reported). Source: benchmarklist.com. Status: years away from saturation. 40 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 5 | 72.3 |
| 2 | Claude Fable 5 | 70.2 |
| 3 | Claude Opus 4.8 | 69.7 |
| 4 | DeepSeek V4 Pro (0813) | 61.1 |
| 5 | Kimi K3 | 58 |
| 6 | GLM-5.3 | 58 |
| 7 | GLM-5.3 Flash | 56.3 |
| 8 | Qwen 3.8 Max | 55.9 |
| 9 | GPT-5.5 | 50.7 |
| 10 | GLM-5.2 | 48.9 |
| 11 | Qwen 3.7 Max | 47.2 |
| 12 | Hy3 | 45.6 |
| 13 | Kimi K2.6 (Thinking) | 42.8 |
| 14 | GLM-5.1 | 42.7 |
| 15 | MiniMax-M3 | 42.1 |
Interactive version: theaggregate.ai/benchmark?slug=nl2repo · How It Works · Data refreshed daily, snapshot 2026-09-05.