kth8/

gpt-oss-20b-GPQA-Diamond-benchmark

Benchmark of openai/gpt-oss-20b against fingertap/GPQA-Diamond dataset. Results are averaged over 4 runs to reduce variance. Accuracy: 64.4% with Python tool.

Metric Value

Correct 510

Incorrect 280

Errors 2

Total samples 792

Python tool calls 306

Python tool errors 13

Total completion tokens 2,525,774

Raw stats: { "accuracy": 0.644, "correct": 510, "incorrect": 280, "error": 2, "total": 792, "python_tool_calls": 306, "python_tool_errors": 13… See the full description on the dataset page: https://huggingface.co/datasets/kth8/gpt-oss-20b-GPQA-Diamond-benchmark.