多模态

只靠文字无法回答、必须看懂图片才能答对的问题。目前只涵盖看图推理，因为各榜只测了这一类。

模型	进入前 10	Artificial Analysis LLM Leaderboard MMMU-Pro	Arena — Vision Overall (style control)
Claude Opus 5.5 Anthropic	2/2	#1 87.7%	#9 1293
GPT 6.1 Sol OpenAI	2/2	#3 86.0%	#6 1294
Gemini 3.7 Flash Google	2/2	#5 85.5%	#5 1297
Claude Fable 5 Anthropic	1/2	—	#1 1308
GPT 6 Astra OpenAI	1/2	#2 86.9%	#16 1286
Qwen 3.8 Max Alibaba	1/2	#11 82.8%	#2 1301
Claude Opus 4.6 Anthropic	1/2	#45 75.4%	#3 1299
Gemini 3.8 Flash Google	1/2	#4 85.6%	#12 1289
Claude Opus 4.7 Anthropic	1/2	#27 78.8%	#4 1298
Claude Opus 5 Anthropic	1/2	#6 84.7%	#13 1287
Gemini 3.5 Flash Google	1/2	#7 84.3%	#19 1285
Muse Spark Meta	1/2	#17 80.5%	#7 1294
GPT 5.6 Sol OpenAI	1/2	#8 83.4%	#14 1287
Muse Spark 1.2 Meta	1/2	—	#8 1293
Gemini 3.6 Flash Google	1/2	#9 83.2%	#21 1281
GPT 6 Sol OpenAI	1/2	#10 82.9%	#30 1270
Muse Spark 1.3 Meta	1/2	#13 82.0%	#10 1290