agents benchmark
NL2Repo Leaderboard
NL2Repo evaluates long-horizon coding capabilities, including repository-level understanding, by requiring models to generate or modify code across entire repositories from natural language specifications.
Updated Sep 5, 2026
Models22
Model coverage22
MetricScore
EvidenceB
NL2Repo Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelDE | Score61.50% | Percentile100.00% | Participants22 | EvidenceC | Evaluated |
| Rank02 | ModelTE | Score58.90% | Percentile95.24% | Participants22 | EvidenceC | Evaluated |
| Rank03 | ModelZA | Score58.00% | Percentile90.48% | Participants22 | EvidenceC | Evaluated |
| Rank04 | ModelDE | Score57.70% | Percentile85.71% | Participants22 | EvidenceC | Evaluated |
| Rank05 | ModelZA | Score56.30% | Percentile80.95% | Participants22 | EvidenceC | Evaluated |
| Rank06 | ModelAC | Score55.90% | Percentile76.19% | Participants22 | EvidenceC | Evaluated |
| Rank07 | ModelDE | Score54.20% | Percentile71.43% | Participants22 | EvidenceC | Evaluated |
| Rank08 | ModelZA | Score48.90% | Percentile66.67% | Participants22 | EvidenceC | Evaluated |
| Rank09 | ModelAC | Score48.10% | Percentile61.90% | Participants22 | EvidenceC | Evaluated |
| Rank10 | ModelAC | Score48.10% | Percentile57.14% | Participants22 | EvidenceC | Evaluated |
| Rank11 | ModelAC | Score47.20% | Percentile52.38% | Participants22 | EvidenceC | Evaluated |
| Rank12 | ModelBY | Score47.00% | Percentile47.62% | Participants22 | EvidenceC | Evaluated |
| Rank13 | ModelTE | Score45.60% | Percentile42.86% | Participants22 | EvidenceC | Evaluated |
| Rank14 | ModelBY | Score43.70% | Percentile38.10% | Participants22 | EvidenceC | Evaluated |
| Rank15 | ModelZA | Score42.70% | Percentile33.33% | Participants22 | EvidenceC | Evaluated |
| Rank16 | ModelAC | Score42.30% | Percentile28.57% | Participants22 | EvidenceC | Evaluated |
| Rank17 | ModelMI | Score42.13% | Percentile23.81% | Participants22 | EvidenceC | Evaluated |
| Rank18 | ModelAC | Score41.10% | Percentile19.05% | Participants22 | EvidenceC | Evaluated |
| Rank19 | ModelMI | Score39.80% | Percentile14.29% | Participants22 | EvidenceC | Evaluated |
| Rank20 | ModelAC | Score37.90% | Percentile9.52% | Participants22 | EvidenceC | Evaluated |
| Rank21 | ModelAC | Score36.20% | Percentile4.76% | Participants22 | EvidenceC | Evaluated |
| Rank22 | ModelAC | Score29.40% | Percentile0.00% | Participants22 | EvidenceC | Evaluated |
NL2Repo Highlights
The leading models and scores on this benchmark.
NL2Repo Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for NL2Repo
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
What is NL2Repo?
What NL2Repo measures and how its scores work.
NL2Repo is an agents benchmark for evaluating coding capabilities across entire repositories from natural language specifications.
It measures long-horizon coding capabilities and repository-level understanding using the Score metric, reported as a ratio.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- NL2Repo
- Modality
- text
- Primary category
- agents
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- overall
LLM Stats. Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about NL2Repo.
Which model scores highest on NL2Repo?
DeepSeek-V4-Pro-0813 is currently ranked first with 61.50%.
What are the top three models on NL2Repo?
The current leaders are DeepSeek-V4-Pro-0813 (61.50%), Hy4 preview (58.90%), and GLM-5.3 (58.00%).
Which NL2Repo model has the lowest official input price?
GLM-5.3-Flash has the lowest matched official input price at $0.08 input / $0.25 output per 1M tokens.
Which models are fastest among NL2Repo results?
The fastest matched records are GLM-5.3 (472.69 tok/s via FriendliAI), GLM-5.3-Flash (74.28 tok/s via FriendliAI), and Qwen3.8 Max (60.87 tok/s via DeepInfra).
Does the highest score result prove overall model quality?
No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.