Sign in to view source links and access this dataset
Description
Evaluation results for the Cogito-405B-thinking model across multiple reasoning benchmarks, published by NousResearch on August 28, 2025. The dataset includes scores for benchmarks such as AIME24, AIME25, ArenaHard, and GPQA Diamond, with metrics like math_pass@1 and winrate. The full description is available on the Hugging Face dataset page.
Use Cases
Benchmarking model reasoning performance based on reported scores for AIME and GPQA Diamond.
Analyzing model consistency across different evaluation tasks based on metrics like winrate and extractive_match.
Comparing the performance of the Cogito-405B model against other models using the provided benchmark results.
Strengths
Includes scores from 8 distinct reasoning benchmarks, such as AIME24 and ArenaHard.
Provides specific metric values, like a 0.910 winrate on ArenaHard and a 0.893 extractive_match on BBH.
Lists overlong rates for some benchmarks, offering insight into response length issues.
Limitations
Description metadata is limited; actual data quality requires manual inspection after download.
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.
Provenance
Source
NousResearch via Hugging Face.
Collection Method
Likely contains aggregated results from running the specified model on standardized evaluation benchmarks.
Freshness
Last updated 2025-08-28 08:37:00; freshness should be verified.
License is unknown; terms of use must be verified before application.