System One Bench

Cloudflare ยท OpenRouter API

Clef

Clef by Cloudflare scores 81.6 on the System One Index (5th of 24) and 78.1 on the Vision Index (6th of 8). Calibration error 2.2%, median latency 310 ms.

System One Index
81.6
5th of 24
English
80.6
Arabic
83.0
Vision Index
78.1
6th of 8
Calibration error
2.2%
Lower is better
Median latency
310 ms
95th percentile 436 ms
Cost per 1,000 calls
$0.116
As billed

Text datasets

DatasetAccuracyRankBest on the boardCalibration error
AG News English90.56th of 2492.55.7%
SST-2 English97.01st of 2497.04.2%
SST-5 English51.514th of 2462.510.9%
DAIR Emotion English58.59th of 2491.518.3%
Banking77 English95.53rd of 2498.54.6%
BoolQ English91.03rd of 2492.56.1%
XNLI English English88.77th of 2493.36.2%
Typed decisions English72.26th of 2479.02.6%
XNLI French French80.010th of 2490.76.2%
XNLI Arabic Arabic78.04th of 2689.37.5%
MASSIVE scenario Arabic88.03rd of 2689.57.5%
MASSIVE intent Arabic89.55th of 2691.55.8%
OSACT4 offensive Arabic89.58th of 2695.52.8%
Tweet sentiment Arabic58.017th of 2680.020.6%
MIRACL rerank Arabic89.03rd of 2690.513.4%
MIRACL pairwise Arabic89.07th of 2691.06.4%

Image datasets

DatasetAccuracyRankBest on the boardCalibration error
SST-2, as an image95.52nd of 896.03.9%
Arabic sentiment, as an image35.06th of 867.036.0%
Oxford-IIIT Pet82.58th of 888.514.8%
VQAv2 yes/no87.54th of 889.52.5%
CountBench90.05th of 894.513.3%

This run

Model id
cloudflare/clef
Runs on
OpenRouter API
Run date
2026-10-07
Latency
API, 8 in flight, timed 15 October 2026
Questions
4,850, 0 failed (counted wrong)

Compare

Ranks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.