벤치마크

분야별 AI 벤치마크

측정 모델이 3개 이상인 벤치마크 310종을 분야별로 묶었습니다(전체 538종 중). 같은 이름이라도 문항·하네스·출처가 다르면 프로필로 나눠 각각 순위를 매깁니다. 측정 모델이 적은 벤치는 각 모델 페이지의 점수표에만 나옵니다.

데이터 기준 2026년 10월 9일

코딩 저장소 수정, 코드 생성, 터미널 작업

33종

벤치마크	대표 프로필	측정 모델	1위	최고점
SWE-bench Verified 프로필 2	LLM Stats 집계	95	Claude Fable 5	95%
SWE-bench Pro 프로필 3	Public 731	88	Claude Fable 5.1	81.2%
Terminal-Bench 2.0 프로필 2	LLM Stats 집계	62	GPT-5.5	82.7%
Terminal-Bench 2.1 프로필 4	LLM Stats 집계	62	DeepSeek-V4.1-Flash	90.6%
SWE-bench Multilingual 프로필 2	LLM Stats 집계	52	Claude Opus 5.5	93.9%
LiveCodeBench v6 프로필 2	LLM Stats 집계	51	Qwen3.8 Flash	91.9%
DeepSWE 1.1	LLM Stats 집계	39	Muse Spark 1.3	75.4%
SciCode 프로필 3	LLM Stats 집계	31	Mistral Large 4	91.8%
DeepSWE 프로필 2	HF 리더보드	28	DeepSeek-V4.1-Flash	74.2%
FrontierCode 1.1 프로필 2	LLM Stats 집계	23	Claude Opus 5.5	54.4%
Terminal-Bench 4.0	LLM Stats 집계	21	Claude Sonnet 5.5	70.6%
IMO-AnswerBench	LLM Stats 집계	18	Nemotron 3 Ultra (550B A55B)	92.3%
FrontierSWE 프로필 2	LLM Stats 집계	17	Claude Fable 5	90%
LiveCodeBench	LLM Stats 집계	13	DeepSeek-V4-Pro-Max	93.5%
Terminal-Bench 3.0 프로필 3	Z.ai 공개 평가	11	GPT-5.6 Sol	34.6%
CodeForces	LLM Stats 집계	10	DeepSeek-V4.1-Flash	3,471
DeepSWE v1.1 프로필 2	OpenAI 공개 평가	10	GPT-5.6 Sol	72.7%
PostTrainBench 프로필 2	LLM Stats 집계	10	GLM-5.3	39.8%
Long-Horizon Terminal Bench	HF 리더보드	9	Kimi-K3	6.0
Terminal-Bench	LLM Stats 집계	9	MiniMax-M2.1	47.9%
SWE-Marathon 1.1	Z.ai 공개 평가	6	Claude Opus 4.8	48.8%
LiveCodeBench Pro	LLM Stats 집계	5	Gemini 3.1 Pro	2,887
SWE Atlas - Codebase QnA	LLM Stats 집계	5	Mistral Large 4	59.4%
SWE-Marathon	LLM Stats 집계	5	GLM-5.3	42.5%
AetherCode	LLM Stats 집계	4	Seed 2.1 Turbo	67.9%
Code Arena	Google 공개 평가	4	Gemini 3.7 Flash	1,588
Multi-SWE-Bench	LLM Stats 집계	4	MiniMax-M2.1	49.4%
FrontierCode	LLM Stats 집계	3	Claude Opus 5	53.4%
FrontierSWE V2	LLM Stats 집계	3	Claude Opus 5.5	62.3%
KernelGen 1P	LLM Stats 집계	3	GPT-5.6 Sol	61.1%
NanoGPT	LLM Stats 집계	3	GPT-5.6 Terra	14.5%
PostTrainBench Lite	LLM Stats 집계	3	GPT-5.6 Terra	51.5%
QwenSWEBench 프로필 2	LLM Stats 집계	3	Qwen3.8 Max	80.7%

에이전트 도구 사용, 브라우징, 컴퓨터 조작

41종

벤치마크	대표 프로필	측정 모델	1위	최고점
BrowseComp 프로필 2	LLM Stats 집계	49	Atria Dawn Preview	92.5%
Toolathlon 프로필 2	LLM Stats 집계	42	GLM-5.3-Flash	78.4%
MCP Atlas · Public 프로필 2	LLM Stats 집계	37	Muse Spark 1.1	88.1%
Claw-Eval 프로필 2	HF 리더보드	29	Ornith-1.5-397B	81.4%
OSWorld-Verified	LLM Stats 집계	26	Qwen3.8 Max	86.1%
Agents' Last Exam 프로필 2	LLM Stats 집계	25	GPT-6 Astra	59.3%
AutomationBench	LLM Stats 집계	24	Mistral Large 4	59.9%
WildClawBench 프로필 2	HF 리더보드	24	dots3-note-prev	61.7%
Finance Agent v2	LLM Stats 집계	21	Ling 3.1 Flash	57.9%
SkillsBench 프로필 2	LLM Stats 집계	19	Qwen3.8 Max	70.2%
OSWorld	LLM Stats 집계	18	Seed 2.1 Pro	78.8%
OSWorld 2.0 프로필 2	LLM Stats 집계	16	Claude Opus 5.5	81.8%
Legal Agent Benchmark	LLM Stats 집계	14	Claude Fable 5	13.3%
APEX Agents	HF 리더보드	13	Kimi-K3	41%
DeepSearch QA 프로필 2	LLM Stats 집계	12	Atria Dawn Preview	96%
GDPval-AA v2 프로필 2	OpenAI 공개 평가	12	Claude Fable 5	1,760
BenchCAD 프로필 2	LLM Stats 집계	11	Claude Sonnet 5.5	74.7%
GDPval-AA	LLM Stats 집계	11	Claude Fable 5.1	1,853
Tau3 Banking 프로필 2	LLM Stats 집계	11	Atria Dawn Preview	41.2%
APEX-Agents	LLM Stats 집계	8	Kimi-K3	37.6%
BrowseComp-zh	LLM Stats 집계	8	Seed 1.8	81.3%
Finance Agent	LLM Stats 집계	8	Claude Opus 4.7	64.4%
MCP-Mark	LLM Stats 집계	8	Kimi-K2.7-Code	81.1%
Agents' Last Exam · CLI	Z.ai 공개 평가	7	GPT-5.6 Sol	28.6%
AutomationBench 1.0.6	Z.ai 공개 평가	7	GLM-5.3	48.2%
BenchCAD (with Python tool)	LLM Stats 집계	7	Claude Sonnet 5.5	96.3%
CoWorkBench 프로필 2	LLM Stats 집계	7	Qwen3.8 Max	74.8%
TAU3-Bench	LLM Stats 집계	7	Qwen3.6 Plus	70.7%
Toolathlon Verified	Z.ai 공개 평가	7	Kimi-K3	76.5%
ClawEval-MM	LLM Stats 집계	6	Qwen3.8 Flash	60.4%
GDPval-AA 2.1	LLM Stats 집계	6	Claude Opus 5.5	1,846
AA-Briefcase Elo	Artificial Analysis	4	Claude Opus 5	1,685
AA-Briefcase Rubric Pass Rate	Artificial Analysis	4	Claude Opus 5	57.2%
AA-Briefcase v1.1	LLM Stats 집계	4	Claude Opus 5.5	1,822
AutomationBench v1.0.6	LLM Stats 집계	4	MiMo-V2.6-Pro	53.1%
GDPval	LLM Stats 집계	4	Seed 2.1 Pro	87.9%
JobBench 프로필 2	LLM Stats 집계	4	MiMo-V2.6-Pro	62%
Toolathlon-Verified	LLM Stats 집계	4	Claude Opus 5.5	77.8%
AA-Briefcase	LLM Stats 집계	3	Kimi-K3	1,548
Management Consulting Tasks (Internal)	LLM Stats 집계	3	GPT-5.6 Sol	43.2%
ZClawBench	LLM Stats 집계	3	Qwen3.7 Max	64.3%

추론·지식 지식, 수학, 추론, 지시 따르기, 장문맥

189종

벤치마크	대표 프로필	측정 모델	1위	최고점
GPQA Diamond 프로필 4	LLM Stats 집계	147	GPT-6 Astra	96%
MMLU-Pro 프로필 2	LLM Stats 집계	101	Sakana Namazu	90.3%
Humanity's Last Exam 프로필 3	LLM Stats 집계	95	Claude Fable 5.1	65%
IFBench 프로필 3	LLM Stats 집계	43	Qwen3.8 Max	82.8%
AIME 2025	LLM Stats 집계	40	Gemini 3 Pro	100%
AIME 2026 프로필 2	LLM Stats 집계	40	GLM-5.2	99.2%
MMMLU	LLM Stats 집계	30	Claude Mythos Preview	92.7%
LVBench	LLM Stats 집계	28	Gemini 3.8 Flash	87.1%
SuperGPQA	LLM Stats 집계	28	Qwen3.7 Max	73.6%
AA-LCR 프로필 3	LLM Stats 집계	27	Solar Mini 4	83.3%
ERQA	LLM Stats 집계	25	Qwen3.8 Max	77.8%
MMLU-Redux	LLM Stats 집계	25	Qwen3.7 Max	95%
IFEval	LLM Stats 집계	24	Qwen3.5-27B	95%
MMLU-ProX	LLM Stats 집계	24	Qwen3.7 Max	87%
NL2Repo 프로필 3	LLM Stats 집계	24	DeepSeek-V4.1-Flash	64%
LiveBench	LLM Stats 집계	23	GPT-5.5	80.7%
BFCL-V4	LLM Stats 집계	22	Atria Dawn Preview	77%
Include	LLM Stats 집계	22	Claude Opus 4.8	87.6%
MathVista-Mini	LLM Stats 집계	22	Kimi-K2.5	90.1%
MMStar	LLM Stats 집계	22	Qwen3.6 Plus	83.3%
MRCR v2 (8-needle)	LLM Stats 집계	22	Muse Spark 1.3	98.5%
HMMT25	LLM Stats 집계	20	Qwen3.6 Plus	94.6%
Tau2 Telecom	LLM Stats 집계	20	Claude Opus 4.6	99.3%
Hallusion Bench	LLM Stats 집계	19	Qwen3.5-27B	70%
HMMT February 2026	HF 리더보드	19	Darwin-180B-RSI	100%
Multi-Challenge	LLM Stats 집계	19	Nova 2 Pro	77.7%
PolyMATH	LLM Stats 집계	19	Qwen3.7 Max	86.5%
t2-bench	LLM Stats 집계	19	Gemini 3.1 Pro	99.3%
AI2D	LLM Stats 집계	18	Qwen3.6 Plus	94.4%
CC-OCR	LLM Stats 집계	18	Qwen3.6 Plus	83.4%
GSM8K	HF 리더보드	18	MiMo-V2.5-Pro	99.6%
OCRBench	LLM Stats 집계	18	Kimi-K2.5	92.3%
MVBench	LLM Stats 집계	16	GLM-5.3-Flash	77.8%
WMT24++	LLM Stats 집계	16	Nemotron 3 Super (120B A12B)	86.7%
BLINK	LLM Stats 집계	15	Seed 2.1 Pro	81.4%
Global PIQA	LLM Stats 집계	15	Gemini 3 Pro	93.4%
HMMT 2025	LLM Stats 집계	15	GPT-5.2	99.4%
MMBench-V1.1	LLM Stats 집계	15	Qwen3.5-122B-A10B	92.8%
ODinW	LLM Stats 집계	15	Qwen3.6 Plus	51.8%
SimpleVQA	LLM Stats 집계	15	Seed 1.8	65.4
WideSearch	LLM Stats 집계	15	Hy4-preview	83.9%
ARC-AGI v2	LLM Stats 집계	14	GPT-6 Astra	95%
LongBench v2	LLM Stats 집계	14	Qwen3.8 Max	66.3%
BFCL-v3	LLM Stats 집계	13	GLM-4.5	77.8%
HMMT Feb 26	LLM Stats 집계	13	Qwen3.7 Max	97.1%
MMLU	LLM Stats 집계	13	Seed 1.8	92.3%
MuirBench	LLM Stats 집계	13	Qwen3 VL 32B Thinking	80.3%
OCRBench-V2 (en)	LLM Stats 집계	13	Qwen3 VL 32B Thinking	68.4%
Program Bench	LLM Stats 집계	13	Claude Opus 5.5	91.2%
ZEROBench	LLM Stats 집계	13	Seed 1.8	11.0
Arena-Hard v2	LLM Stats 집계	12	Qwen3 VL 235B A22B Instruct	77.4%
C-Eval	LLM Stats 집계	12	Qwen3 Max Thinking	93.7%
CharadesSTA	LLM Stats 집계	12	Qwen3 VL 235B A22B Instruct	64.8%
Graphwalks BFS >128k	LLM Stats 집계	12	GPT-5.2	94%
MAXIFE	LLM Stats 집계	12	Qwen3.7 Max	89.2%
Multi-IF	LLM Stats 집계	12	Ling-3.0-flash	87.7%
NOVA-63	LLM Stats 집계	12	Qwen3.5-397B-A17B	59.1%
OfficeQA Pro	LLM Stats 집계	12	Seed 2.1 Pro	72.2%
SimpleQA	LLM Stats 집계	12	Gemini 3 Pro	72.1%
Humanity's Last Exam (no tools, text-only)	LLM Stats 집계	11	Claude Opus 5.5	64.4%
InfoVQAtest	LLM Stats 집계	11	Kimi-K2.5	92.6%
MathArena Apex	LLM Stats 집계	11	DeepSeek-V4-Pro-Max	90.2%
MedXpertQA	LLM Stats 집계	11	Muse Spark	78.4%
OCRBench-V2 (zh)	LLM Stats 집계	11	Qwen3 VL 235B A22B Thinking	63.5%
VITA-Bench	LLM Stats 집계	11	Qwen3.5-397B-A17B	49.7%
WritingBench	LLM Stats 집계	11	Qwen3 VL 235B A22B Thinking	86.7%
DeepPlanning	LLM Stats 집계	10	Qwen3.7-Plus	62.3%
DocVQAtest	LLM Stats 집계	10	Qwen3 VL 235B A22B Instruct	97.1%
Humanity's Last Exam (with tools, text-only)	LLM Stats 집계	10	Claude Opus 5.5	67.7%
LiveBench 20241125	LLM Stats 집계	10	Qwen3 VL 235B A22B Thinking	79.6%
MLVU	LLM Stats 집계	10	Qwen3.7-Plus	87.4%
MM-MT-Bench	LLM Stats 집계	10	Qwen3 VL 235B A22B Instruct	8.5
Tau2 Retail	LLM Stats 집계	10	Claude Opus 4.6	91.9%
DynaMath	LLM Stats 집계	9	Qwen3.6 Plus	88%
AndroidWorld_SR	LLM Stats 집계	8	Qwen3.5-35B-A3B	71.1%
ARC-AGI-3 프로필 2	OpenAI 공개 평가	8	GPT-5.6 Sol	7.8%
CountBench	LLM Stats 집계	8	Seed 1.8	96.3
Creative Writing v3	LLM Stats 집계	8	Qwen3 VL 235B A22B Instruct	86.5%
EmbSpatialBench	LLM Stats 집계	8	Qwen3.6-27B	0.8
Graphwalks parents >128k	LLM Stats 집계	8	Claude Opus 4.6	95.4%
Job Bench	LLM Stats 집계	8	Muse Spark 1.3	64.9%
MLVU-M	LLM Stats 집계	8	Qwen3 VL 32B Instruct	82.1%
RefCOCO-avg	LLM Stats 집계	8	Qwen3.6 Plus	0.9
RefSpatialBench	LLM Stats 집계	8	Seed 1.8	56.3
Tau2 Airline	LLM Stats 집계	8	Nova 2 Omni	68.8%
AndroidWorld	LLM Stats 집계	7	Qwen3.8 Max	85.3%
ARC-AGI	LLM Stats 집계	7	GPT-6 Astra	98.5%
Artificial Analysis	LLM Stats 집계	7	GPT-5.6 Sol	59%
BioMysteryBench	LLM Stats 집계	7	Claude Opus 5	90.1%
FrontierMath	LLM Stats 집계	7	GPT-5.6 Sol	89%
FrontierMath Tier 1-3 v2	OpenAI 공개 평가	7	GPT-5.6 Sol	89%
HLE · with tools	Z.ai 공개 평가	7	GPT-5.6 Sol	64.5%
MMVU	LLM Stats 집계	7	GLM-5.3-Flash	80.5%
OCRBench_V2	LLM Stats 집계	7	Qwen3.7-Plus	67.1%
Seal-0	LLM Stats 집계	7	Kimi-K2.5	57.4%
BIG-Bench Extra Hard	LLM Stats 집계	6	Gemma 4 31B	74.4%
CritPT	LLM Stats 집계	6	Hy4-preview	16.9%
MMLongBench-Doc	LLM Stats 집계	6	Qwen3.6 Plus	0.6
ProgramBench · Almost Solved	Z.ai 공개 평가	6	Claude Fable 5	33%
V*	LLM Stats 집계	6	Kimi-K2.6	96.9%
VLMsAreBlind	LLM Stats 집계	6	Qwen3.5-35B-A3B	97%
WorldVQA	LLM Stats 집계	6	Qwen3.7-Plus	61.1%
ZEROBench-Sub	LLM Stats 집계	6	Qwen3.5-122B-A10B	0.4
ArXivMath	LLM Stats 집계	5	Claude Opus 5.5	91.2%
Beyond AIME	LLM Stats 집계	5	Seed 2.1 Turbo	88%
Chartography	LLM Stats 집계	5	Claude Sonnet 5.5	90.2%
MotionBench	LLM Stats 집계	5	Seed 2.1 Pro	74.9%
OJBench	LLM Stats 집계	5	Kimi-K2.6	60.6%
SuperChem	LLM Stats 집계	5	Hy4-preview	66.4%
TVBench	LLM Stats 집계	5	Seed 2.1 Pro	80.5%
AA Intelligence Index	Google 공개 평가	4	GPT-5.6 Terra	57.0
Artifacts Bench	LLM Stats 집계	4	Ling-3.0-flash	77%
ChartQAPro	LLM Stats 집계	4	Seed 2.1 Pro	70.9%
CSimpleQA	LLM Stats 집계	4	DeepSeek-V4-Pro-Max	84.4%
DocVQA	LLM Stats 집계	4	Qwen2.5-VL-72B-Instruct	96.4%
DRACO	LLM Stats 집계	4	Ling 3.1 Flash	85.5%
DUDE	LLM Stats 집계	4	Seed 2.1 Turbo	83.1%
EMMA	LLM Stats 집계	4	Seed 2.1 Pro	79.3%
FrontierMath Tier 4 (v2)	LLM Stats 집계	4	GPT-6 Astra	97.6%
HorizonMath	LLM Stats 집계	4	Hy4-preview	8.8%
Hypersim	LLM Stats 집계	4	Qwen3.5-35B-A3B	0.1
Intelligence Index v4.1.1	Artificial Analysis	4	Claude Opus 5	63.1
LingoQA	LLM Stats 집계	4	Qwen3.7-Plus	83.4%
LiveSports-3K	LLM Stats 집계	4	Seed 1.8	77.5%
MathVista	LLM Stats 집계	4	Seed 2.1 Pro	90.7%
Minerva	LLM Stats 집계	4	Seed 2.1 Pro	70.7%
MRCR v2 · 512K–1M	OpenAI 공개 평가	4	GPT-5.5	74%
MRCR v2 (8-needle, 512K-1M)	LLM Stats 집계	4	Muse Spark 1.3	98.1%
OneMillion Bench	LLM Stats 집계	4	Seed 2.1 Pro	68.8%
OVBench	LLM Stats 집계	4	Seed 2.1 Pro	70%
OVOBench	LLM Stats 집계	4	Seed 2.1 Pro	80.7%
PinchBench	LLM Stats 집계	4	Nemotron 3 Ultra (550B A55B)	90%
ProfBench	LLM Stats 집계	4	Nemotron 3 Ultra (550B A55B)	56%
SimpleQA Verified	LLM Stats 집계	4	Inkling	43.9%
SUNRGBD	LLM Stats 집계	4	Qwen3.5-122B-A10B	0.4
TempCompass	LLM Stats 집계	4	Seed 1.8	86.9%
TIR-Bench	LLM Stats 집계	4	Qwen3.6 Plus	61.6%
TOMATO	LLM Stats 집계	4	Seed 2.1 Pro	79.5%
Vending-Bench 2	LLM Stats 집계	4	Claude Opus 4.6	801,759%
VisFactor	LLM Stats 집계	4	Seed 2.1 Pro	51.4%
VisuLogic	LLM Stats 집계	4	Seed 2.1 Pro	0.5
VLMsAreBiased	LLM Stats 집계	4	Seed 2.1 Pro	83.6%
Workspace Bench	LLM Stats 집계	4	Qwen3.8 Max	67.7%
AA-Index	LLM Stats 집계	3	GLM-4.5	67.7%
AITZ_EM	LLM Stats 집계	3	Qwen2.5-VL-72B-Instruct	83.2%
Android Control High_EM	LLM Stats 집계	3	Qwen2.5 VL 32B Instruct	69.6%
Android Control Low_EM	LLM Stats 집계	3	Qwen2.5-VL-72B-Instruct	93.7%
ArcAGI2	LLM Stats 집계	3	Seed 2.1 Pro	62.5%
Big Finance Bench	LLM Stats 집계	3	GPT-5.6 Sol	53%
Bird-SQL (dev)	LLM Stats 집계	3	IBM Granite 4.2 30B	41.9%
Capture-the-Flag Challenges (Internal)	LLM Stats 집계	3	GPT-5.6 Sol	96.7%
ChartQA	LLM Stats 집계	3	Qwen2.5-VL-72B-Instruct	89.5%
CL-bench	LLM Stats 집계	3	Hy3	23.8%
COLLIE	LLM Stats 집계	3	Mistral Medium 3.5	95.8%
Connectors	LLM Stats 집계	3	GPT-5.6 Sol	100%
ContPhy	LLM Stats 집계	3	Seed 2.1 Pro	63.6%
CorpusQA 1M	LLM Stats 집계	3	DeepSeek-V4-Pro-Max	62%
CrossVid	LLM Stats 집계	3	Seed 2.1 Pro	65%
DSBench-Hard	LLM Stats 집계	3	DeepSeek-V4-Pro-0813	67.2%
FullStackBench en	LLM Stats 집계	3	Qwen3.5-122B-A10B	62.6%
FullStackBench zh	LLM Stats 집계	3	Qwen3.5-122B-A10B	58.7%
Global-MMLU	LLM Stats 집계	3	Claude Opus 5.5	94.3%
Graphwalks BFS 1M	LLM Stats 집계	3	GPT-5.6 Sol	77.1%
HLE-Verified	LLM Stats 집계	3	Gemini 3.8 Flash	54.9%
Humanity's Last Exam (with tools)	LLM Stats 집계	3	Claude Sonnet 5.5	64.5%
Internal Research Debugging Evaluation	LLM Stats 집계	3	GPT-5.6 Sol	68.3%
LatchBio SingleCellBench	LLM Stats 집계	3	Claude Opus 5.5	61.2%
LatchBio SpatialBench Verified	LLM Stats 집계	3	Claude Sonnet 5.5	72.5%
MILU	LLM Stats 집계	3	Claude Opus 5.5	93.1%
MLS-Bench Lite	LLM Stats 집계	3	Kimi-K3	48.3%
MMAU	LLM Stats 집계	3	Inkling	77.2%
MobileWorld	LLM Stats 집계	3	Qwen3.8 Max	77.8%
MRCR 1M	LLM Stats 집계	3	DeepSeek-V4-Pro-Max	83.5%
Nuscene	LLM Stats 집계	3	Qwen3.5-122B-A10B	15.4%
OfficeQA	LLM Stats 집계	3	Claude Opus 5.5	78.9%
PaperBench	LLM Stats 집계	3	Qwen3.8 Max	93%
PHYBench	LLM Stats 집계	3	Hy3	77.4%
PMC-VQA	LLM Stats 집계	3	Qwen3.5-122B-A10B	63.3%
QVHighlights	LLM Stats 집계	3	Nova 2 Lite	77.2%
RealKIE-FCC	LLM Stats 집계	3	Nova 2 Pro	67%
RecreationBench	LLM Stats 집계	3	Qwen3.8 Flash	49.9%
RSI Index	LLM Stats 집계	3	GPT-5.6 Sol	57.9%
RULER 128k	LLM Stats 집계	3	IBM Granite 4.2 30B	81.4%
RULER 64k	LLM Stats 집계	3	IBM Granite 4.2 30B	90.0%
Search and Function-Calling	LLM Stats 집계	3	GPT-5.6 Terra	94.6%
SlakeVQA	LLM Stats 집계	3	Qwen3.5-122B-A10B	81.6%
SpreadSheetBench-v1	LLM Stats 집계	3	MiniMax-M3	89.4%
TreeBench	LLM Stats 집계	3	Seed 2.1 Pro	71.1%
USAMO 2026	LLM Stats 집계	3	MiniMax-M3	36.0

멀티모달 이미지, 영상, 화면, 문서 이해

27종

벤치마크	대표 프로필	측정 모델	1위	최고점
MMMU-Pro 프로필 3	LLM Stats 집계	69	Gemini 3.5 Flash	83.6%
CharXiv-R	LLM Stats 집계	48	Claude Mythos Preview	93.2%
ScreenSpot-Pro 프로필 2	LLM Stats 집계	44	GPT-6 Astra	92.7%
MathVision	LLM Stats 집계	34	Kimi-K3	97.8%
RealWorldQA 프로필 2	LLM Stats 집계	25	Qwen3.8 Flash	88.5%
VideoMMMU	LLM Stats 집계	23	Gemini 3 Pro	87.6%
OmniDocBench 1.5 프로필 3	LLM Stats 집계	21	MiniMax-M3	91.6%
ScreenSpot	LLM Stats 집계	16	Qwen3 VL 32B Instruct	95.8%
BabyVision	LLM Stats 집계	13	DeepSeek-V4.1-Flash	89.6%
MMMU	LLM Stats 집계	13	Qwen3.6 Plus	86%
CharXiv-D	LLM Stats 집계	12	Seed 2.1 Pro	95.5%
ExtractBench	HF 리더보드	11	Darwin-180B-RSI-R3	90.3%
gdp.pdf 프로필 2	LLM Stats 집계	11	Claude Sonnet 5	81.6%
Video-MME	LLM Stats 집계	11	Seed 2.1 Pro	89.2%
VideoMME w sub.	LLM Stats 집계	10	Qwen3.8 Max	90.4%
VideoMME w/o sub.	LLM Stats 집계	10	Qwen3.5-122B-A10B	83.9%
MMMU (val)	LLM Stats 집계	9	Qwen3 VL 32B Thinking	78.1%
SWE-Bench Multimodal	LLM Stats 집계	7	Claude Opus 5.5	61.4%
LongVideoBench	LLM Stats 집계	6	Seed 2.1 Pro	80.6%
MMMU-Pro · no tools	OpenAI 공개 평가	5	GPT-5.6 Sol	83%
MMMU-Pro (with tools)	LLM Stats 집계	4	GPT-5.6 Sol	84.6%
Video-MME v2	HF 리더보드	4	MiniMax-M3	85.4%
VideoSimpleQA	LLM Stats 집계	4	Seed 2.1 Pro	76.4%
Vision2Web	LLM Stats 집계	4	Qwen3.8 Max	69%
CharXiv Reasoning · without CI	Qwen 공개 평가	3	Qwen3.8-27B	83.7%
MMBench-Video	LLM Stats 집계	3	Qwen2.5-VL-72B-Instruct	2.0%
VideoHolmes	LLM Stats 집계	3	Seed 2.1 Pro	68.2%

과학·보안 과학·의료·보안 전문 평가

18종

벤치마크	대표 프로필	측정 모델	1위	최고점
CyberGym 프로필 2	LLM Stats 집계	23	MiMo-V2.6-Flash	95.1%
HealthBench Professional 프로필 2	LLM Stats 집계	15	Claude Sonnet 5.5	69.2%
ExploitBench 프로필 3	LLM Stats 집계	14	GPT-6 Astra	100%
HealthBench	LLM Stats 집계	10	Claude Sonnet 5.5	65.4%
ExploitGym	LLM Stats 집계	8	GPT-6 Astra	42.4%
SEC-Bench Pro 프로필 2	LLM Stats 집계	8	GPT-6 Astra	85.4%
AA-Omniscience Index 프로필 2	Artificial Analysis	7	Claude Opus 5	37.1
GeneBench Pro	OpenAI 공개 평가	7	GPT-5.6 Sol	28.7%
HealthBench Hard	LLM Stats 집계	6	Muse Spark	42.8%
LifeSciBench 프로필 2	OpenAI 공개 평가	6	GPT-5.6 Sol	59.9%
FrontierScience Research	LLM Stats 집계	5	Muse Spark	38.3%
Terminal-Bench-Science 0.1	LLM Stats 집계	5	GPT-6 Astra	64.6%
AA-Omniscience Accuracy	Artificial Analysis	4	Claude Opus 5	60.9%
FrontierScience Olympiad	LLM Stats 집계	4	Seed 2.1 Turbo	76%
GeneBench-Pro	LLM Stats 집계	4	GPT-6 Astra	37.8%
HealthBench Consensus	LLM Stats 집계	4	GPT-5.6 Sol	95.5%
MedChemBench (Internal)	LLM Stats 집계	3	GPT-5.6 Sol	48.3%
OmniScience	LLM Stats 집계	3	Nemotron 3 Ultra (550B A55B)	78.7%

속도·비용 API 속도, 첫 토큰 지연

2종

벤치마크	대표 프로필	측정 모델	1위	최고점
Artificial Analysis Output Speed	Artificial Analysis	4	Gemini 3.7 Flash	324 tok/s
Artificial Analysis Time to First Chunk	Artificial Analysis	4	Gemini 3.7 Flash	9.06초