MeasureBench Synthetic: leaderboard

Each model reads the value and unit off 1,170 programmatically rendered images of 39 instrument designs with known readings. The score is the percentage of answers whose value falls in the interval and whose unit matches; it is the rendered counterpart of the MeasureBench real-world photographs.

Metric: Accuracy (%). Source: raw.githubusercontent.com. Saturation forecast: Around 2029. 13 models tracked. Updated 17 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3 VL 235B A22B Instruct19#540
2GPT-5 Mini17.9#359
3GPT-516.9#177
4Claude Opus 4.1Non-reasoning13.3#279
5InternVL3.5-38BNon-reasoning12.6#1017
6Qwen 2.5 VL 72B Instruct11.7#849
7Claude Sonnet 411#411
8Qwen 2.5 VL 7B Instruct10.9#1893
9Qwen 2.5 VL 32B Instruct10.5#1074
10Mistral Medium 3.18.5#865
11InternVL3.5-8BNon-reasoning7.7#1588
12Mistral Small 3.26.5#1645
13Grok 46.2#315

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=measurebench-synthetic · How It Works · Data refreshed daily, snapshot 2026-11-17.