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.
Reasoning benchmark
BIG-Bench Hard. Every model the catalog carries a published BBH value for, ranked by that value.
A suite of 23 challenging tasks from the BIG-Bench collaborative benchmark where prior language models failed to exceed average human performance, even with chain-of-thought prompting. 9 models in the ModelScale catalog have a published value.
Each point is a model with a published BBH value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
9 models with a published BBH 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 | Mixed reasoning tasks |
|---|---|---|---|
| 1 | Anthropic | 86.6 | |
| 2 | SPSoofi S 30B-A3B | Soofi Project | 78.8 |
| 3 | OPMiniCPM5-1B | OpenBMB | 71.89 |
What the benchmark measures and how its value is published, from the source’s own record.
BBH (BIG-Bench Hard) is a reasoning benchmark published by Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them. A suite of 23 challenging tasks from the BIG-Bench collaborative benchmark where prior language models failed to exceed average human performance, even with chain-of-thought prompting. Its evaluation set is described by the source as 23 tasks. The source rates its difficulty as advanced reasoning.
Each model's result is published by Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them as mixed reasoning tasks. ModelScale prints that value exactly as published, and never converts or rescales it. This page ranks the 9 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.
| Microsoft |
| 69 |
| 5 | IBM | 60.37 |
| 6 | IBM | 59.68 |
| 7 | 53 |
| 8 | IBM | 33.28 |
| 9 | IBM | 33.07 |
Evidence key: ObservedLast good
Rows are ordered by the value Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them 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.