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 52.6 #6 of 339
Knowledge 5
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 49.9% #22 of 323+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 15.1% #5 of 323
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA 42.4 #5 of 323
SimpleQA Verifiedv1 Accuracy on short factual questions with verified answers, answered without web access 76% #2 of 84
Mystery Game Puzzles Models choose the best move in puzzle positions from an undisclosed game Epoch 64% #5 of 65
Reasoning 6
ARC-AGI-1public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 98% #10 of 64+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 98.5% #1 of 67
ARC-AGI-2public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 92.1% #10 of 66+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 87.1% #10 of 69
ARC-AGI-3semi-private Measures interactive game solving through relative human action efficiency ARC 46.3% #3 of 19
Chess Puzzles Models choose the best next move from a novel chess position Epoch 59% #6 of 69
Math 8
AA Analyst Agent Data-analyst tasks: the model explores data with tools and reports findings AA 48.8% #10 of 43
FrontierMath Tier 4 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 78% #9 of 55
FrontierMath Tiers 1–3 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 86% #10 of 70
MathArenaArXivMath · August 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 82.5% #6 of 14+4 versions & slices ArXivMath · September 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 87.7% #2 of 9BrokenArXiv · August 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 54.2% #8 of 13BrokenArXiv · September 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 49.7% #4 of 9ArXivLean · September 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 39.6% #2 of 6
Science 5
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 57.1% #3 of 339
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 27.1% #16 of 326
Vals Science Index Scientific problem solving across the index’s component evaluations Vals 55.4% #5 of 19
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 61.8% #4 of 142
Terminal-Bench Science0.1 Agents complete expert-authored research workflows across five scientific domains in a terminal AA 52.9% #6 of 45
Coding 8
Terminal-Bench4.0 Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite 57.1% #4 of 134
FrontierCode1.1 Agents implement changes in real repositories, graded against maintainer-defined requirements AA 43.5% #13 of 38
WebDev Arena Human pairwise preferences in the stated webdev vote pool Arena 1,678 #9 of 97+2 versions & slices October 23, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,674 #10 of 98October 19, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,678 #9 of 97
SWE-bench Pro (v2 full) Agents resolve repository issues under the network-locked v2 protocol; full and 51-task Hard results are separate Scale 99.2% #2 of 16
SWE-bench Pro (v2 hard) Agents resolve repository issues under the network-locked v2 protocol; full and 51-task Hard results are separate Scale 92.2% #3 of 16
MirrorCode Agents reimplement software projects by observing their behavior Epoch 58.9% #4 of 10
Agents 6 Leads on 2
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 77.5% #1 of 135
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 1,483 #22 of 142
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 1,631 #17 of 177
AutomationBenchZapier run End-to-end completion of business workflows using real SaaS applications Zapier 50.1% #1 of 2
EBR Bench Agents learn to play Earthborne Rangers across repeated playthroughs Epoch 64.8% #3 of 24
Furniture Assembly Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step Epoch 72.5% #5 of 31
Multimodal 1 Leads on 1
Chartography Expert chart reasoning over professional scientific, medical, financial and engineering charts Surge 71.6% #1 of 37
Writing 17
Arena Textcreative writing · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,523 #2 of 197+13 versions & slices creative writing Human pairwise preferences in the stated text vote pool Arena 1,516 #2 of 195overall Human pairwise preferences in the stated text vote pool Arena 1,524 #1 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,600 #1 of 198English Human pairwise preferences in the stated text vote pool Arena 1,526 #1 of 205French Human pairwise preferences in the stated text vote pool Arena 1,543 #1 of 164instruction following Human pairwise preferences in the stated text vote pool Arena 1,528 #2 of 205non-English Human pairwise preferences in the stated text vote pool Arena 1,519 #1 of 205Polish Human pairwise preferences in the stated text vote pool Arena 1,530 #1 of 142Russian Human pairwise preferences in the stated text vote pool Arena 1,537 #1 of 203Spanish Human pairwise preferences in the stated text vote pool Arena 1,508 #5 of 168overall · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,532 #1 of 188creative writing · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,527 #2 of 187overall · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,526 #1 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 80.9% #14 of 96
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 1,758 #29 of 94
Lech Mazur Short-Story Judges compare stories written to ten required elements; the comparison score can be negative Lech 0.7 #15 of 45
Health & law 6
Harvey LAB-AA1.1 Legal agents work from case documents; every rubric criterion must pass and material hallucinations receive no credit AA 3.9% #10 of 23
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 13.9% #40 of 103
Vals LegalBench Legal reasoning and legal-language understanding on LegalBench Vals 88.3% #3 of 89
Vals MedCode Medical coding accuracy on patient encounters Vals 58.8% #3 of 78
Vals MedScribe Accuracy of clinical notes produced from consultation transcripts Vals 87.4% #16 of 80
Vals Legal Research Bench Grounded legal research; a task passes only if all criteria pass Vals 54.8% #4 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 79.7% #50 of 317
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 21.8% #22 of 135
Legacy · saturated or no longer reported 2
GPQA DiamondEpoch Models answer graduate-level science multiple-choice questions Epoch 96.4% #1 of 134
OTIS Mock AIME 20242025 Models solve the stated OTIS mock AIME mathematics problems Epoch 100% #1 of 131
No benchmark matches
OpenDeepSeek V4.1 Flash 1017high-4%-87%$0.30 / $1.20No PDF input+5
CheaperGLM 5.3 FlashmaxOpen-8%-94%$0.28Defaults to max · Self-hosted+4
Kimi K3maxOpen-4%+50%$2.00Sampling parameters are fixed+3
Muse Spark OpenxhighOpen-10%-50%$0.84Reasoning off leaves output unchanged · Self-hosted / llama.cpp+1