agents benchmark
NL2Repo
NL2Repo evaluates long-horizon coding capabilities including repository-level understanding, where models must generate or modify code across entire repositories from natural language specifications.
Updated Aug 7, 2026
Published models14
Registry coverage14
MetricScore
EvidenceB
NL2Repo leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 55.9% | 100.0% | 14 | C | |
| 2 | DE | 54.2% | 92.3% | 14 | C | |
| 3 | ZA | 48.9% | 84.6% | 14 | C | |
| 4 | AC | 47.2% | 76.9% | 14 | C | |
| 5 | BY | 47.0% | 69.2% | 14 | C | |
| 6 | TE | 45.6% | 61.5% | 14 | C | |
| 7 | BY | 43.7% | 53.9% | 14 | C | |
| 8 | ZA | 42.7% | 46.1% | 14 | C | |
| 9 | MI | 42.1% | 38.5% | 14 | C | |
| 10 | AC | 41.1% | 30.8% | 14 | C | |
| 11 | MI | 39.8% | 23.1% | 14 | C | |
| 12 | AC | 37.9% | 15.4% | 14 | C | |
| 13 | AC | 36.2% | 7.7% | 14 | C | |
| 14 | AC | 29.4% | 0.0% | 14 | C |
Score distribution
Top published rows on the benchmark's original scale.
NL2Repo highlights
The top published results on this benchmark's own scale.
What is NL2Repo?
Definition and scoring fields from the benchmark registry.
NL2Repo evaluates long-horizon coding capabilities including repository-level understanding, where models must generate or modify code across entire repositories from natural language specifications.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- NL2Repo
- Modality
- text
- Primary category
- agents
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- nl2repo|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about NL2Repo.
Which model scores highest on NL2Repo?
Qwen3.8 Max is currently ranked first with 55.9%.
What does NL2Repo measure?
NL2Repo evaluates long-horizon coding capabilities including repository-level understanding, where models must generate or modify code across entire repositories from natural language specifications.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
14 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.