CommonsenseQA
Five-choice questions derived from ConceptNet relations, designed to require everyday knowledge beyond a supplied passage.
KEAR reached 89.4% on the hidden test set versus the original 88.9% human reference in 2021; the benchmark's authors then introduced the adversarially collected CommonsenseQA 2.0.
Performance Timeline
Longitudinal progression of model scores against human baselines.Performance & Historical Trajectory
Empirical score progression across model release dates and evaluation rounds.
| Model | Score | Date | Source Type | Provenance |
|---|---|---|---|---|
| GPT-4o | 85.4% | 2024-05-13 | vendor-reported | Source ↗ |
| Llama 3 70B (7-shot) | 83.8% | 2024-04-18 | vendor-reported | Source ↗ |
| CPACE | 89.8% | 2023-05-14 | independent | Source ↗ |
| KEAR ensemble | 89.4% | 2021-12-06 | vendor-reported | Source ↗ |
| KEAR single model | 86.1% | 2021-12-06 | vendor-reported | Source ↗ |
| DeBERTa-xxlarge (1.5B) | 83.8% | 2021-01-06 | independent | Source ↗ |
| RoBERTa-large | 72.1% | 2019-07-25 | independent | Source ↗ |
| BERT-Large | 55.9% | 2018-11-02 | independent | Source ↗ |
Human Baseline & Difficulty Horizon
Calibrated human reference points, specialist benchmarks, and ceiling thresholds.Human crowdworker accuracy on multiple-choice commonsense knowledge triples.
Metric & Scoring Methodology
Verification protocols, aggregation formulas, and specialized metric variants.accuracy (%)Dataset & Compute Cost
Evaluation volume, public availability, API pricing, and local hardware requirements.$5 – $20 USD for full benchmark evaluation run on frontier APIs.
1x NVIDIA RTX 4090 (24GB) or A100 (40GB/80GB) via vLLM / SGLang
How to Run & Reproduce
Standardized evaluation protocols, CLI commands, and reproducible runner templates.lm_eval --model hf --model_args pretrained=<model_path> --tasks commonsenseqa --batch_size autoopencompass --datasets commonsenseqa --models <model_config># Standard API Evaluation Loop
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": prompt}],
temperature=0.0,
)When publishing results for CommonsenseQA, always report the exact prompt template, few-shot exemplar ordering, sampling temperature (temperature=0), maximum reasoning budget tokens, and the precise timestamped model snapshot ID.
Contamination & Memorization Analysis
Audit of pretraining exposure risks, memorization vectors, and refresh policies.Static fixed snapshot
Public on web / HuggingFace