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
DeepSeek DSBench Hard. Every model the catalog carries a published DSBench-Hard value for, ranked by that value.
DeepSeek's internal hard coding-agent benchmark. 3 models in the ModelScale catalog have a published value.
Each point is a model with a published DSBench-Hard value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
3 models with a published DSBench-Hard 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 | Provider-reported score |
|---|---|---|---|
| 1 | DeepSeek | 67.2 | |
| 1 | DeepSeek | 67.2 | |
| 3 | DeepSeek | 59.6 |
What the benchmark measures and how its value is published, from the source’s own record.
DSBench-Hard (DeepSeek DSBench Hard) is a coding benchmark published by DeepSeek V4 Flash 0731 update. DeepSeek's internal hard coding-agent benchmark. Its evaluation set is described by the source as internal hard coding-agent tasks. The source rates its difficulty as advanced coding-agent challenges.
Each model's result is published by DeepSeek V4 Flash 0731 update as provider-reported score. ModelScale prints that value to 1 decimal place, exactly as published, and never converts or rescales it. This page ranks the 3 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.
Evidence key: ObservedLast good
Rows are ordered by the value DeepSeek V4 Flash 0731 update 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.