SpecVQA: leaderboard

SpecVQA (2026) asks questions about 620 scientific spectra (NMR, IR, XRD, Raman, mass, UV-Vis and XPS) from published papers, with expert-revised descriptive and reasoning questions in English and Chinese; GPT-o4-mini judges each answer against the reference, numeric answers within 5 percent. This board is the unweighted mean accuracy of the four language-and-level splits, in percent; higher is better.

Metric: Accuracy (0-1, times 100), the unweighted mean of the English and Chinese descriptive and reasoning splits; expert-revised question-answer pairs on 620 spectra (NMR, IR, XRD, Raman, MS, UV-Vis, XPS) from published papers; each answer judged correct or not by GPT-o4-mini against the reference, numeric answers within a 5 percent tolerance; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Flash (Preview)78.72#156
2Gemini 3 Pro (Preview)77.87#123
3Gemini 2.5 Pro76.15#280
4Gemini 2.5 Flash72.74#466
5GPT-5High70.59#185
6O4 Mini (2025-04-16)70.54#375
7O3 (2025-04-16)70.31#238
8GPT-5.1Non-reasoning65.78#248
9GPT-5.2Non-reasoning65.4#190
10Claude Sonnet 4.5Thinking59.41#267
11Doubao-Seed-1.6-FlashThinking58.67#1268
12Claude Sonnet 4Thinking56.88#414
13Qwen 3 VL 8B (Thinking)56.31#924
14Doubao-Seed-1.6Thinking56.02
15Qwen 3 VL 8B Instruct51.58#1043

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

Interactive version: theaggregate.ai/benchmark?slug=specvqa · How It Works · Data refreshed daily, snapshot 2026-11-22.