InfiniteBench
100k+ token benchmark mixing realistic and synthetic tasks across English, Chinese, code, math, and retrieval.
InfiniteBench remains valuable for 100k+ token stress testing, but its public static split and newer complementary suites make it better classified as nearing-saturation than as the freshest active frontier signal.
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-4 (paper Table 3 Average) | 45.63% | 2024-02-21 | independent | Source ↗ |
| YaRN-Mistral (paper Table 3 Average) | 19.96% | 2024-02-21 | independent | Source ↗ |
| Kimi-Chat (paper Table 3 Average) | 34.73% | 2024-02-21 | independent | Source ↗ |
| Claude 2 (paper Table 3 Average) | 37.06% | 2024-02-21 | independent | Source ↗ |
Human Baseline & Difficulty Horizon
Calibrated human reference points, specialist benchmarks, and ceiling thresholds.Human reading comprehension accuracy across 100K+ token synthetic and document retrieval tasks.
Metric & Scoring Methodology
Verification protocols, aggregation formulas, and specialized metric variants.task-specific accuracy / ROUGE aggregateDataset & 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 infinitebench --batch_size autoopencompass --datasets infinitebench --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 InfiniteBench, 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