Benchmark By Score Rank
Overall 1
AA Intelligence Index4.3.2 AA Intelligence Index 4.3.2: Artificial Analysis combines tests of knowledge, reasoning, coding and tool use into one score AA 25 #84 of 337
Knowledge 4
AA OmniscienceAccuracy AA Omniscience · Accuracy: Measures correct answers on the AA Omniscience factual knowledge test AA 41.6% #42 of 321+1 versions & slices Hallucination Rate AA Omniscience · Hallucination Rate: Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 67.7% #127 of 321
AA Omniscience Index AA Omniscience Index: Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA 2 #60 of 321
SimpleQA Verifiedv1 SimpleQA Verified v1: Accuracy on short factual questions with verified answers, answered without web access 40.3% #42 of 83
Reasoning 5
ARC-AGI-1public evaluation ARC-AGI-1 · public-eval: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 90.6% #30 of 61+1 versions & slices semi-private ARC-AGI-1 · semi-private: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 79.5% #36 of 64
ARC-AGI-2public evaluation ARC-AGI-2 · public-eval: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 37.8% #33 of 63+1 versions & slices semi-private ARC-AGI-2 · semi-private: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 36.5% #36 of 66
Chess Puzzles Chess Puzzles: Models choose the best next move from a novel chess position Epoch 21% #49 of 68
Math 3
AA Analyst Agent AA Analyst Agent: Data-analyst tasks: the model explores data with tools and reports findings AA 23.8% #29 of 43
FrontierMath Tier 4 (v2) FrontierMath Tier 4 (v2): Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 4.9% #51 of 54
FrontierMath Tiers 1–3 (v2) FrontierMath Tiers 1–3 (v2): Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 33.3% #55 of 69
Science 4
Humanity's Last Exam Humanity's Last Exam: Very hard expert questions across many fields, written to stump today’s models AA 31.9% #74 of 337
CritPt (physics) CritPt (physics): Research-level physics problems that need many careful reasoning steps AA 5.4% #77 of 324
SciCode SciCode: Writing code for real scientific computing problems (physics, biology, math) AA 47% #66 of 140
Terminal-Bench Science0.1 Terminal-Bench Science 0.1: Agents complete expert-authored research workflows across five scientific domains in a terminal AA 0% #27 of 33
Coding 8
Terminal-Bench4.0 Terminal-Bench 4.0 · source setting 2: Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite 1% #54 of 132+1 versions & slices 2.1 Terminal-Bench 2.1 · source setting 2: Older, easier version of Terminal-Bench; Still widely reported AA 55.1% #62 of 156
DeepSWE1.1 DeepSWE 1.1: Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition 5% #38 of 44
FrontierCode1.1 FrontierCode 1.1: Agents implement changes in real repositories, graded against maintainer-defined requirements AA 15.8% #26 of 28
WebDev Arena WebDev Arena: Human pairwise preferences in the stated webdev vote pool Arena 1,414 #64 of 97+1 versions & slices full cohort WebDev Arena · snapshot 2026-10-19 · native board settings: Human pairwise preferences in the stated webdev vote pool Arena 1,414 #64 of 97
SWE-bench Pro (v2 full) SWE-bench Pro (v2 full): Agents resolve repository issues under the network-locked v2 protocol; full and 51-task Hard results are separate Scale 89.9% #15 of 16
SWE-bench Pro (v2 hard) SWE-bench Pro (v2 hard): Agents resolve repository issues under the network-locked v2 protocol; full and 51-task Hard results are separate Scale 56.9% #14 of 16
Agents 6
AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 5% #84 of 133
AA-Briefcase1.1 · Elo AA-Briefcase 1.1 · Elo: Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 832 #73 of 140
EnterpriseOps-Gym-AA EnterpriseOps-Gym-AA: Agents complete stateful enterprise workflows in eight domains; final database state determines success AA 38% #26 of 43
GDPval-AA (work tasks)2.1 GDPval-AA (work tasks) 2.1: Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 28.8 #69 of 175
τ³-benchbanking · 1.0.1 τ-bench official · 3-banking-1.0.1: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools Sierra 25% #17 of 21
τ³-BankingAA τ³-Banking · AA: Agents retrieve banking knowledge and execute tools to resolve customer-support workflows AA 29.1% #36 of 134
Multimodal 1
MMMU-Pro (vision) MMMU-Pro (vision): College-level questions that require understanding images: charts, diagrams, photos AA 73.5% #60 of 127
Writing 17
Arena Textcreative writing Arena Text · creative writing: Human pairwise preferences in the stated text vote pool Arena 1,381 #83 of 195+14 versions & slices overall Arena Text · overall: Human pairwise preferences in the stated text vote pool Arena 1,441 #63 of 196Chinese Arena Text · chinese: Human pairwise preferences in the stated text vote pool Arena 1,492 #48 of 181English Arena Text · english: Human pairwise preferences in the stated text vote pool Arena 1,442 #60 of 188French Arena Text · french: Human pairwise preferences in the stated text vote pool Arena 1,454 #53 of 149German Arena Text · german: Human pairwise preferences in the stated text vote pool Arena 1,447 #40 of 154instruction following Arena Text · instruction following: Human pairwise preferences in the stated text vote pool Arena 1,425 #57 of 188Japanese Arena Text · japanese: Human pairwise preferences in the stated text vote pool Arena 1,426 #30 of 138Korean Arena Text · korean: Human pairwise preferences in the stated text vote pool Arena 1,403 #39 of 144non-English Arena Text · non english: Human pairwise preferences in the stated text vote pool Arena 1,433 #41 of 188Polish Arena Text · polish: Human pairwise preferences in the stated text vote pool Arena 1,453 #25 of 133Russian Arena Text · russian: Human pairwise preferences in the stated text vote pool Arena 1,429 #53 of 186Spanish Arena Text · spanish: Human pairwise preferences in the stated text vote pool Arena 1,447 #45 of 155overall · full cohort Arena Text · overall · snapshot 2026-10-15 · native board settings: Human pairwise preferences in the stated text vote pool Arena 1,441 #49 of 188creative writing · full cohort Arena Text · creative writing · snapshot 2026-10-15 · native board settings: Human pairwise preferences in the stated text vote pool Arena 1,387 #70 of 187
EQ-Bench Longform Creative Writing1.11 Longform Creative Writing 1.11: Models plan, revise and write a story across eight turns, graded with a writing rubric EQ-Bench 72.5% #34 of 95
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 1,611 #40 of 91
Health & law 6
Harvey LAB-AA1.1 Harvey LAB-AA 1.1: Legal agents work from case documents; every rubric criterion must pass and material hallucinations receive no credit AA 0% #18 of 23
MLCR-AA MLCR-AA: Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 12.2% #37 of 64
Vals LegalBench Vals LegalBench: Legal reasoning and legal-language understanding on LegalBench Vals 83% #49 of 89
Vals MedCode Vals MedCode: Medical coding accuracy on patient encounters Vals 41.2% #46 of 78
Vals MedScribe Vals MedScribe: Accuracy of clinical notes produced from consultation transcripts Vals 85.4% #25 of 80
Vals Legal Research Bench Vals Legal Research Bench: Grounded legal research; a task passes only if all criteria pass Vals 28.4% #43 of 69
Documents 2
AA-LCR (long context)1.1 AA-LCR (long context) 1.1: Questions that need reading and connecting facts across very long documents (around 100K tokens) AA 77.3% #76 of 275
GDP.pdfAA all-pass GDP.pdf · AA all-pass: Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 12.8% #46 of 133
Legacy · saturated or no longer reported 3
GPQA Diamond GPQA Diamond: Graduate-level biology, chemistry and physics questions that can’t be solved by a quick web search 87.2% #57 of 315+1 versions & slices Epoch GPQA Diamond · source setting 2: Models answer graduate-level science multiple-choice questions Epoch 88.3% #43 of 133
OTIS Mock AIME 20242025 OTIS Mock AIME 2024 2025: Models solve the stated OTIS mock AIME mathematics problems Epoch 88.9% #49 of 130
No benchmark matches
CheaperMiMo v2.6 ProthinkingOpen+85%-68%$0.13Memory (4-bit): 571 GB → 615 GB+4
BetterGPT-6 Lunamedium+20%-88%$0.017Function calling only with none · Chat Completions+3
CheaperDeepSeek V4.1 Flash 1017maxOpen+80%-70%$0.26Pass back reasoning_content · Chat Completions+4
CheaperStep 5 PreviewthinkingOpen+75%-17%$1.03web_search not available · Chat Completions+2
CheaperClaude Haiku 5.5max+74%-88%$0.21Remove sampling parameters+5
MiniMax M3.1 Flashthinking+71%-85%$0.22Complete model response must be appended
GLM 5.3maxOpen+79%+25%$2.01No image input+4