SimpleQA
4,326 short fact-seeking questions chosen to stump GPT-4 — a hallucination test graded correct / incorrect / not-attempted.
Still discriminating hard. Frontier models remain well short of the ceiling on parametric recall: GPT-4o 38% correct at launch, GPT-4.5 ~62% by early 2025 — climbing but nowhere near solved, and reasoning-tuned models do not automatically help (o1-mini scored 8%). What would move it off 'active': frontier 'correct' crossing ~85–90% (approaching the label ceiling), OR retrieval-augmented deployment becoming universal enough that closed-book parametric recall stops mattering. Neither has happened.
Performance Timeline
Longitudinal progression of model scores against human baselines.Performance & Historical Trajectory
Empirical score progression across model release dates and evaluation rounds.
Human Baseline & Difficulty Horizon
Calibrated human reference points, specialist benchmarks, and ceiling thresholds.Consensus agreement rate among independent third-party human fact-checkers on 4,326 factual queries.
Metric & Scoring Methodology
Verification protocols, aggregation formulas, and specialized metric variants.correct (accuracy over all questions, %)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 simpleqa --batch_size autoopencompass --datasets simpleqa --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 SimpleQA, 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