Benchmark By Score Rank
Overall 1
AA Intelligence Index4.3.2 Artificial Analysis combines tests of knowledge, reasoning, coding and tool use into one score AA 34.2 #45 of 339
Knowledge 5
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 51.3% #18 of 323+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 50.7% #92 of 323
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA 26.5 #19 of 323
SimpleQA Verifiedv1 Accuracy on short factual questions with verified answers, answered without web access 64.4% #12 of 84
Mystery Game Puzzles Models choose the best move in puzzle positions from an undisclosed game Epoch 56% #9 of 65
Reasoning 6
ARC-AGI-1public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 79.3% #38 of 64+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 72.2% #42 of 67
ARC-AGI-2public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 25.6% #38 of 66+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 31.5% #40 of 69
ARC-AGI-3semi-private Measures interactive game solving through relative human action efficiency ARC 0.1% #19 of 19
Chess Puzzles Models choose the best next move from a novel chess position Epoch 56% #11 of 69
Math 9
FrontierMath Erdos Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 1.5% #5 of 13
FrontierMath Tier 4 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 90% #5 of 55
FrontierMath Tiers 1–3 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 89.8% #5 of 70
MathArenaArXivMath · August 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 93% #2 of 14+5 versions & slices ArXivMath · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 91% #3 of 24ArXivMath · May 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 90% #3 of 19BrokenArXiv · August 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 78.6% #4 of 13BrokenArXiv · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 95.4% #4 of 24BrokenArXiv · May 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 87% #4 of 18
Science 5
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 34.9% #65 of 339
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 16.3% #43 of 326
Vals Science Index Scientific problem solving across the index’s component evaluations Vals 45% #7 of 19
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 50.2% #51 of 142
Terminal-Bench Science0.1 Agents complete expert-authored research workflows across five scientific domains in a terminal AA 3.3% #28 of 45
Coding 6
Terminal-Bench4.0 Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite 9.1% #41 of 134
DeepSWE1.1 Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition 70.5% #7 of 44
ITBench SRE IT incident tasks: diagnose and fix broken systems like a site-reliability engineer AA 49.4% #7 of 37
WebDev Arena Human pairwise preferences in the stated webdev vote pool Arena 1,683 #8 of 97+2 versions & slices October 23, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,681 #9 of 98October 19, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,683 #8 of 97
Agents 8
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 53.9% #33 of 135
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 886 #68 of 142
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 1,196 #50 of 177
EBR Bench Agents learn to play Earthborne Rangers across repeated playthroughs Epoch 53.3% #6 of 24
Furniture Assembly Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step Epoch 58.3% #8 of 31
SpireClimb Agents play Slay the Spire under the named game and difficulty setting Epoch 7.5% #3 of 7
SpireClimb StS1 A20 Heart Agents play Slay the Spire under the named game and difficulty setting Epoch 1.7% #3 of 7
SpireClimb StS2 A10 Agents play Slay the Spire under the named game and difficulty setting Epoch 13.3% #3 of 7
Multimodal 3
MMMU-Pro (vision) College-level questions that require understanding images: charts, diagrams, photos AA 80.2% #20 of 128
Arena Vision Human pairwise preferences in the stated vision vote pool Arena 1,240 #51 of 88
Chartography Expert chart reasoning over professional scientific, medical, financial and engineering charts Surge 53.6% #4 of 37
Writing 19
Arena Textcreative writing · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,377 #81 of 197+16 versions & slices creative writing Human pairwise preferences in the stated text vote pool Arena 1,438 #38 of 195overall Human pairwise preferences in the stated text vote pool Arena 1,459 #41 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,412 #110 of 198English Human pairwise preferences in the stated text vote pool Arena 1,396 #105 of 205French Human pairwise preferences in the stated text vote pool Arena 1,395 #104 of 164German Human pairwise preferences in the stated text vote pool Arena 1,373 #90 of 167instruction following Human pairwise preferences in the stated text vote pool Arena 1,411 #74 of 205Japanese Human pairwise preferences in the stated text vote pool Arena 1,369 #72 of 152Korean Human pairwise preferences in the stated text vote pool Arena 1,333 #92 of 156non-English Human pairwise preferences in the stated text vote pool Arena 1,388 #93 of 205Polish Human pairwise preferences in the stated text vote pool Arena 1,380 #87 of 142Russian Human pairwise preferences in the stated text vote pool Arena 1,403 #83 of 203Spanish Human pairwise preferences in the stated text vote pool Arena 1,387 #102 of 168overall · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,398 #91 of 188creative writing · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,380 #77 of 187overall · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,398 #95 of 198
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 82.8% #6 of 96
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 2,125 #4 of 94
Health & law 5
Harvey LAB-AA1.1 Legal agents work from case documents; every rubric criterion must pass and material hallucinations receive no credit AA 3.6% #11 of 23
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 25.6% #14 of 103
Vals MedCode Medical coding accuracy on patient encounters Vals 47.1% #29 of 78
Vals MedScribe Accuracy of clinical notes produced from consultation transcripts Vals 82% #44 of 80
Vals Legal Research Bench Grounded legal research; a task passes only if all criteria pass Vals 28.9% #44 of 73
Documents 2
AA-LCR (long context)1.1 Questions that need reading and connecting facts across very long documents (around 100K tokens) AA 79.3% #54 of 317
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 24.2% #12 of 135
Legacy · saturated or no longer reported 2
GPQA DiamondEpoch Models answer graduate-level science multiple-choice questions Epoch 94.3% #8 of 134
OTIS Mock AIME 20242025 Models solve the stated OTIS mock AIME mathematics problems Epoch 100% #1 of 131
No benchmark matches
OpenKimi K3max-9%+1,406%$2.00Context: 1.05M → 1M+4
CheaperGemini 3.8 Flashlow-23%-63%*$0.75 / $3.75Context: 1.05M → 1M+2
Claude Haiku 5.5xhigh-25%-7%$0.12Context: 1.05M → 1M+6
MiMo v2.6 ProthinkingOpen-21%same$0.13Context: 1.05M → 1M+4
Muse Spark OpenxhighOpen-23%+536%$0.84Context: 1.05M → 1M+2
DeepSeek V4.1 Flash 1017maxOpen-29%+97%$0.26Context: 1.05M → 1M+5
Qwen3.8 27BxhighOpen-32%+763%$1.15Context: 1.05M → 262K+5