agents benchmark
Ramp SWE-Bench v2 (Bash) is a private coding benchmark with 500 backend engineering tasks based on pull requests merged into production after engineer review.
Updated Nov 22, 2026
Higher score ranks better on this benchmark.
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelOP | Score75.00% | Percentile100.00% | Participants22 | EvidenceB | Evaluated |
| Rank02 | ModelGO | Score74.00% | Percentile95.24% | Participants22 | EvidenceB | Evaluated |
| Rank03 | ModelAN | Score73.00% | Percentile90.48% | Participants22 | EvidenceB | Evaluated |
| Rank04 | ModelOP | Score72.40% | Percentile85.71% | Participants22 | EvidenceB | Evaluated |
| Rank05 | ModelAN | Score71.20% | Percentile80.95% | Participants22 | EvidenceB | Evaluated |
| Rank06 | ModelGO | Score70.80% | Percentile76.19% | Participants22 | EvidenceB | Evaluated |
| Rank07 | ModelAN | Score63.80% | Percentile71.43% | Participants22 | EvidenceB | Evaluated |
| Rank08 | ModelOP | Score60.60% | Percentile66.67% | Participants22 | EvidenceB | Evaluated |
| Rank09 | ModelME | Score57.80% | Percentile61.90% | Participants22 | EvidenceB | Evaluated |
| Rank10 | ModelAN | Score57.60% | Percentile57.14% | Participants22 | EvidenceB | Evaluated |
| Rank11 | ModelXA | Score55.40% | Percentile52.38% | Participants22 | EvidenceB | Evaluated |
| Rank12 | ModelGO | Score54.80% | Percentile47.62% | Participants22 | EvidenceB | Evaluated |
| Rank13 | ModelOP | Score54.60% | Percentile42.86% | Participants22 | EvidenceB | Evaluated |
| Rank14 | ModelME | Score54.40% | Percentile38.10% | Participants22 | EvidenceB | Evaluated |
| Rank15 | ModelDE | Score53.40% | Percentile33.33% | Participants22 | EvidenceB | Evaluated |
| Rank16 | ModelAN | Score52.80% | Percentile28.57% | Participants22 | EvidenceB | Evaluated |
| Rank17 | ModelOP | Score52.00% | Percentile23.81% | Participants22 | EvidenceB | Evaluated |
| Rank18 | ModelGO | Score47.00% | Percentile19.05% | Participants22 | EvidenceB | Evaluated |
| Rank19 | ModelMA | Score45.40% | Percentile14.29% | Participants22 | EvidenceB | Evaluated |
| Rank20 | ModelOP | Score42.60% | Percentile9.52% | Participants22 | EvidenceB | Evaluated |
| Rank21 | ModelRA | Score34.00% | Percentile4.76% | Participants22 | EvidenceB | Evaluated |
| Rank22 | ModelNV | Score12.00% | Percentile0.00% | Participants22 | EvidenceB | Evaluated |
The leading models and scores on this benchmark.
A closer view of the leading scores on this benchmark.
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
What Ramp SWE-Bench v2 (Bash) measures and how its scores work.
Each model receives a one-shot prompt in the same sandboxed copy of Ramp's codebase and produces a review-ready patch using the bash-only mini-swe-agent harness.
It measures Score as pass@1 ratio at high reasoning effort with web tools disabled; a task passes when an LLM judge finds no material difference where the engineer-reviewed merged solution is better.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
LLM Stats. Benchmark scores retain their original unit. Overall score eligibility is shown separately.
Common questions about Ramp SWE-Bench v2 (Bash).
GPT-6.2 Astra is currently ranked first with 75.00%.
The current leaders are GPT-6.2 Astra (75.00%), Gemini 4 Carbon (74.00%), and Claude Fable 5.2 (73.00%).
Muse Ember has the lowest matched official input price at $0.10 input / $0.50 output per 1M tokens.
The fastest matched records are Nemotron 3.5 Super VL (120B A12B) (96.29 tok/s via DeepInfra), Muse Ember (63.51 tok/s via Meta Model API), and Claude Sonnet 5.5 (42.00 tok/s via Anthropic).
No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.
It measures Score as pass@1 ratio at high reasoning effort with web tools disabled; a task passes when an LLM judge finds no material difference where the engineer-reviewed merged solution is better.
Yes. Higher values rank better for this benchmark.
22 model results are currently shown.
No. This benchmark is shown for reference but does not contribute to the overall score.
Ranking basisThis ramp swe-bench v2 (bash) AI model leaderboard uses descending score in the benchmark original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.
Selection summary
GPT-6.2 Astra currently leads Ramp SWE-Bench v2 (Bash) with 75.00%. It is the top model on this specific benchmark, while the best LLM for the broader task should also be checked against other benchmarks, price, and runtime.
Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.