AgentIF
707 instructions from 50 real-world agents, averaging 1,723 words and 11.9 constraints — whether instruction following survives deployment-scale prompts.
Moved to the site's nearing-saturation bucket: the benchmark is still useful for exposing agentic constraint failures, but the public static set is small, top CSR is already near 60%, and the main value is shifting from broad frontier ranking to targeted failure analysis.
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.Human annotator constraint compliance rate across multi-turn agent system prompt interactions.
Metric & Scoring Methodology
Verification protocols, aggregation formulas, and specialized metric variants.CSR (Constraint Success Rate, %)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 agentif --batch_size autoopencompass --datasets agentif --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 AgentIF, 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