reasoning benchmark
BigCodeBench-Hard
BigCodeBench-Hard is a subset of 148 challenging programming tasks from BigCodeBench, designed to evaluate large language models' ability to solve complex, real-world programming problems. These tasks require diverse function calls from multiple libraries across 7 domains including computation, networking, data analysis, and visualization. The benchmark tests compositional reasoning and the ability to implement complex instructions that span 139 libraries with an average of 2.8 libraries per task.
Updated Aug 7, 2026
Published models1
Registry coverage1
MetricScore
EvidenceB
BigCodeBench-Hard leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 27.0% | 100.0% | 1 | C |
BigCodeBench-Hard highlights
The top published results on this benchmark's own scale.
What is BigCodeBench-Hard?
Definition and scoring fields from the benchmark registry.
BigCodeBench-Hard is a subset of 148 challenging programming tasks from BigCodeBench, designed to evaluate large language models' ability to solve complex, real-world programming problems. These tasks require diverse function calls from multiple libraries across 7 domains including computation, networking, data analysis, and visualization. The benchmark tests compositional reasoning and the ability to implement complex instructions that span 139 libraries with an average of 2.8 libraries per task.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- BigCodeBench-Hard
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- bigcodebench-hard|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about BigCodeBench-Hard.
Which model scores highest on BigCodeBench-Hard?
Qwen2.5-Coder 32B Instruct is currently ranked first with 27.0%.
What does BigCodeBench-Hard measure?
BigCodeBench-Hard is a subset of 148 challenging programming tasks from BigCodeBench, designed to evaluate large language models' ability to solve complex, real-world programming problems. These tasks require diverse function calls from multiple libraries across 7 domains including computation, networking, data analysis, and visualization. The benchmark tests compositional reasoning and the ability to implement complex instructions that span 139 libraries with an average of 2.8 libraries per task.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
1 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.