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 46.4 #15 of 339
Knowledge 5
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 49.7% #23 of 323+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 29.3% #31 of 323
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA 32 #9 of 323
SimpleQA Verifiedv1 Accuracy on short factual questions with verified answers, answered without web access 59.4% #16 of 84
Mystery Game Puzzles Models choose the best move in puzzle positions from an undisclosed game Epoch 29% #26 of 65
Reasoning 6
ARC-AGI-1public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 95% #25 of 64+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 90% #27 of 67
ARC-AGI-2public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 57.5% #31 of 66+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 58.8% #32 of 69
ARC-AGI-3semi-private Measures interactive game solving through relative human action efficiency ARC 1.7% #10 of 19
Chess Puzzles Models choose the best next move from a novel chess position Epoch 41% #29 of 69
Math 12
AA Analyst Agent Data-analyst tasks: the model explores data with tools and reports findings AA 55% #4 of 43
FrontierMath Erdos Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 0% #6 of 13
FrontierMath Tier 4 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 17.1% #43 of 55
FrontierMath Tiers 1–3 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 53% #45 of 70
MathArenaArXivMath · August 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 46.5% #9 of 14+7 versions & slices ArXivMath · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 70.1% #12 of 24ArXivMath · May 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 60.8% #11 of 19ArXivMath · September 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 45.3% #7 of 9BrokenArXiv · August 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 35.1% #9 of 13BrokenArXiv · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 52.8% #10 of 24BrokenArXiv · May 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 40% #9 of 18BrokenArXiv · September 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 27.7% #7 of 9
Science 5
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 43.1% #32 of 339
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 18% #36 of 326
Vals Science Index Scientific problem solving across the index’s component evaluations Vals 34.8% #12 of 19
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 57.8% #16 of 142
Terminal-Bench Science0.1 Agents complete expert-authored research workflows across five scientific domains in a terminal AA 9.5% #19 of 45
Coding 7
Terminal-Bench4.0 Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite 25.8% #26 of 134
DeepSWE1.1 Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition 33.3% #30 of 44
FrontierCode1.1 Agents implement changes in real repositories, graded against maintainer-defined requirements AA 39.6% #21 of 38
ITBench SRE IT incident tasks: diagnose and fix broken systems like a site-reliability engineer AA 42.1% #15 of 37
WebDev Arena Human pairwise preferences in the stated webdev vote pool Arena 1,637 #15 of 97+2 versions & slices October 23, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,638 #16 of 98October 19, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,637 #15 of 97
Agents 4
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 65.6% #12 of 135
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 1,638 #7 of 142
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 1,722 #8 of 177
Furniture Assembly Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step Epoch 20.8% #29 of 31
Multimodal 3
MMMU-Pro (vision) College-level questions that require understanding images: charts, diagrams, photos AA 78.2% #34 of 128
Arena Vision Human pairwise preferences in the stated vision vote pool Arena 1,231 #54 of 88
Chartography Expert chart reasoning over professional scientific, medical, financial and engineering charts Surge 14.7% #31 of 37
Writing 20
Arena Textcreative writing · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,381 #78 of 197+16 versions & slices creative writing Human pairwise preferences in the stated text vote pool Arena 1,424 #49 of 195overall Human pairwise preferences in the stated text vote pool Arena 1,444 #58 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,445 #87 of 198English Human pairwise preferences in the stated text vote pool Arena 1,411 #96 of 205French Human pairwise preferences in the stated text vote pool Arena 1,431 #81 of 164German Human pairwise preferences in the stated text vote pool Arena 1,372 #91 of 167instruction following Human pairwise preferences in the stated text vote pool Arena 1,408 #76 of 205Japanese Human pairwise preferences in the stated text vote pool Arena 1,356 #76 of 152Korean Human pairwise preferences in the stated text vote pool Arena 1,378 #64 of 156non-English Human pairwise preferences in the stated text vote pool Arena 1,385 #98 of 205Polish Human pairwise preferences in the stated text vote pool Arena 1,412 #66 of 142Russian Human pairwise preferences in the stated text vote pool Arena 1,399 #88 of 203Spanish Human pairwise preferences in the stated text vote pool Arena 1,403 #88 of 168overall · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,401 #90 of 188creative writing · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,387 #71 of 187overall · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,398 #96 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 83% #4 of 96
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 2,012 #8 of 94
Lech Mazur Short-Story Judges compare stories written to ten required elements; the comparison score can be negative Lech 0.4 #21 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 9.4% #3 of 23
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 21.1% #18 of 103
Vals LegalBench Legal reasoning and legal-language understanding on LegalBench Vals 84.4% #29 of 89
Vals MedCode Medical coding accuracy on patient encounters Vals 49.6% #18 of 78
Vals MedScribe Accuracy of clinical notes produced from consultation transcripts Vals 89.4% #7 of 80
Vals Legal Research Bench Grounded legal research; a task passes only if all criteria pass Vals 47.1% #11 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 80% #44 of 317
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 23.2% #15 of 135
Legacy · saturated or no longer reported 2
GPQA DiamondEpoch Models answer graduate-level science multiple-choice questions Epoch 92.7% #18 of 134
OTIS Mock AIME 20242025 Models solve the stated OTIS mock AIME mathematics problems Epoch 98.1% #24 of 131
No benchmark matches
CheaperGemini 3.8 Flashhigh+9%-50%*$3.49Sampling parameters are ignored+1
BetterGPT-6.1 Sollow+4%+33%$0.13Remove unsupported parameters+5
OpenDeepSeek V4.1 Flash 1017high+5%-83%$0.30 / $1.20Pass back reasoning_content · Chat Completions+4
CheaperGLM 5.3 FlashmaxOpen+1%-92%$0.28Defaults to max · Self-hosted+4
Kimi K3maxOpen+5%+100%$2.00Sampling parameters are fixed+3
Muse Spark OpenxhighOpen-1%-33%$0.84Reasoning off leaves output unchanged · Self-hosted / llama.cpp+1