Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Coding benchmark
Every model the catalog carries a published SWE-Rebench value for, ranked by that value.
A continuously updated software engineering benchmark by Nebius using fresh GitHub issues to avoid contamination. Models are evaluated 5 times per problem under a fixed ReAct scaffolding; the Resolved Rate (best pass@1) is reported. 13 models in the ModelScale catalog have a published value.
Each point is a model with a published SWE-Rebench value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
13 models with a published SWE-Rebench value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Code patch generation |
|---|---|---|---|
| 1 | Anthropic | 65.3 | |
| 2 | Z.AI | 62.8 | |
| 3 | Z.AI | 62.7 | |
What the benchmark measures and how its value is published, from the source’s own record.
SWE-Rebench is a coding benchmark published by SWE-Rebench: Contamination-Free Evaluation of Software Engineering Agents. A continuously updated software engineering benchmark by Nebius using fresh GitHub issues to avoid contamination. Models are evaluated 5 times per problem under a fixed ReAct scaffolding; the Resolved Rate (best pass@1) is reported. Its evaluation set is described by the source as fresh GitHub issues (rolling window). The source rates its difficulty as professional software engineering.
Each model's result is published by SWE-Rebench: Contamination-Free Evaluation of Software Engineering Agents as code patch generation. ModelScale prints that value exactly as published, and never converts or rescales it. This page ranks the 13 models that carry a published value, highest value first; a model the source never scored is left out rather than ranked at zero. The source does not state whether a higher value is the better result, so neither does this page.
| DeepSeek |
| 60.9 |
| 5 | Anthropic | 60.7 |
| 6 | Alibaba | 58.9 |
| 7 | Z.AI | 58.7 |
| 8 | Moonshot AI | 58.5 |
| 9 | OpenAI | 58.2 |
| 10 | Cursor | 58 |
| 11 | Alibaba | 53.7 |
| 12 | MiniMax | 51.9 |
| 13 | 41.6 |
Evidence key: Observed
Rows are ordered by the value SWE-Rebench: Contamination-Free Evaluation of Software Engineering Agents published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.