MathematicsSaturated
Trust Score:
25

GSM8K

8.5K grade-school math word problems (2–8 arithmetic steps) — the default elementary-math benchmark and the classic chain-of-thought demonstration.

Launched: Refresh: static
Status Assessment (saturated):

Frontier models cleared ~95% by 2023–24 and the reasoning-model era pushed scores to ~97–99%, collapsing headroom. But saturation here is the sharpest 'saturated ≠ solved' case on the wiki: GSM-Symbolic (Apple, 2024) showed that changing only names/numbers, and especially adding one irrelevant clause (GSM-NoOp), drops accuracy heavily across models — so the SCORE is saturated while the CAPABILITY (robust arithmetic reasoning) is not. saturated_date 2024-01 marks the point frontier scores stopped discriminating; the discriminating signal moved to harder math (MATH → AIME → FrontierMath) and to robustness variants.

Performance Timeline

Longitudinal progression of model scores against human baselines.

Performance & Historical Trajectory

Empirical score progression across model release dates and evaluation rounds.

Independent Vendor-Reported Human Baseline (96.0%)
ModelScoreDateSource TypeProvenance
GPT-4 (5-shot CoT)92%2023-03-14vendor-reportedSource ↗
PaLM 540B (CoT + self-consistency)74.4%2022-03-21independentSource ↗
PaLM 540B (8-shot CoT)56.9%2022-01-28independentSource ↗
GPT-3 175B (fine-tuned)35%2021-10-27vendor-reportedSource ↗

Human Baseline & Difficulty Horizon

Calibrated human reference points, specialist benchmarks, and ceiling thresholds.
Measured Human Score96.0%Domain Expert Baseline
Baseline Protocol & Interpretation

OpenAI measured human problem solvers scoring 96% due to occasional arithmetic and reading slip-ups (unverified crowdworkers score ~80%).

Metric & Scoring Methodology

Verification protocols, aggregation formulas, and specialized metric variants.
Primary Metric:accuracy — exact match on the final numeric answer (%)
Scoring Engine:exact-match

Dataset & Compute Cost

Evaluation volume, public availability, API pricing, and local hardware requirements.
Total Dataset Size1,319Annotated evaluation items
Public Test SetPublicOpenly mirrored on repositories
Access GatingOpen AccessUnrestricted download
Evaluation LicenseOpen AccessDataset usage and redistribution terms
Frontier API Compute Cost:

$5 – $20 USD for full benchmark evaluation run on frontier APIs.

Recommended Local GPU Setup:

1x NVIDIA RTX 4090 (24GB) or A100 (40GB/80GB) via vLLM / SGLang

Official Dataset & Benchmark Files:Download / View Dataset Repository ↗

How to Run & Reproduce

Standardized evaluation protocols, CLI commands, and reproducible runner templates.
Prompt Regimezero-shot
Reasoning Modedirect
Sampling Temp0
Pass@k Budgetk = 1
Tools & SandboxPure Text
Scoring Verifierexact-match
Option AEleutherAI LM-Evaluation-Harness (Open-Weight Models)
lm_eval --model hf --model_args pretrained=<model_path> --tasks gsm8k --batch_size auto
Option BOpenCompass Evaluation Framework
opencompass --datasets gsm8k --models <model_config>
Python APIDeterministic Inference Loop Snippet
# 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,
)
Standardized Reporting Requirement:

When publishing results for GSM8K, 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.
Overall Contamination Risk:HIGH
Refresh Cadence:

Static fixed snapshot

Test Set Exposure:

Public on web / HuggingFace