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 36 #44 of 339
Knowledge 6
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 49.1% #26 of 323+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 91.9% #276 of 323
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA 0.8 #67 of 323
EQ-Bench3 Judge-rated responses to social and emotionally nuanced scenarios, scored through the source pairwise ranking EQ-Bench 1,555 #8 of 53
SimpleQA Verifiedv1 Accuracy on short factual questions with verified answers, answered without web access 52.9% #22 of 84
Mystery Game Puzzles Models choose the best move in puzzle positions from an undisclosed game Epoch 43% #15 of 65
Reasoning 5
ARC-AGI-1public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 95.5% #25 of 64+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 90.5% #25 of 67
ARC-AGI-2public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 58.9% #30 of 66+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 56.3% #33 of 69
Chess Puzzles Models choose the best next move from a novel chess position Epoch 46% #21 of 69
Math 14
MathArenaAIME 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 96.7% #6 of 24+11 versions & slices HMMT February 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 93.9% #9 of 24Apex Answer accuracy on the named mathematics competition under the recorded effort MathArena 28.1% #9 of 34Apex Shortlist Answer accuracy on the named mathematics competition under the recorded effort MathArena 87.8% #6 of 30ArXivMath · April 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 51.3% #12 of 17ArXivMath · February 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 51.6% #6 of 20ArXivMath · January 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 73.9% #2 of 18ArXivMath · March 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 55.8% #7 of 14BrokenArXiv · April 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 22.1% #10 of 15BrokenArXiv · February 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 13.3% #7 of 15BrokenArXiv · March 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 15.2% #6 of 13USAMO 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 60.7% #4 of 8
FrontierMath Tier 4 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 26.8% #33 of 55
FrontierMath Tiers 1–3 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 64.6% #33 of 70
Science 4
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 41% #42 of 339
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 18% #36 of 326
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 51% #49 of 142
Terminal-Bench Science0.1 Agents complete expert-authored research workflows across five scientific domains in a terminal AA 5.7% #24 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 14.1% #34 of 134+1 versions & slices 2.1 Older, easier version of Terminal-Bench; Still widely reported AA 78.7% #24 of 156
DeepSWE1.1 Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition 62.8% #19 of 44+1 versions & slices 1 Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition Datacurve 7.5% #16 of 20
FrontierCode1.1 Agents implement changes in real repositories, graded against maintainer-defined requirements AA 38.8% #22 of 38
WebDev Arena Human pairwise preferences in the stated webdev vote pool Arena 1,582 #29 of 97+2 versions & slices October 23, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,582 #30 of 98October 19, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,582 #29 of 97
Agents 6
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 56.7% #29 of 135
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 1,255 #42 of 142
EnterpriseOps-Gym-AA Agents complete stateful enterprise workflows in eight domains; final database state determines success AA 49.6% #2 of 43
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 1,458 #35 of 177
Toolathlon-Verified Toolathlon-Verified as reported in this launch evaluation; the post is the authority for the recorded task and grading scope 74.4% #5 of 6
τ³-BankingAA Agents retrieve banking knowledge and execute tools to resolve customer-support workflows AA 39.6% #17 of 134
Writing 20
Arena Textcreative writing · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,446 #26 of 197+16 versions & slices creative writing Human pairwise preferences in the stated text vote pool Arena 1,446 #30 of 195overall Human pairwise preferences in the stated text vote pool Arena 1,465 #36 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,481 #55 of 198English Human pairwise preferences in the stated text vote pool Arena 1,459 #34 of 205French Human pairwise preferences in the stated text vote pool Arena 1,473 #29 of 164German Human pairwise preferences in the stated text vote pool Arena 1,458 #28 of 167instruction following Human pairwise preferences in the stated text vote pool Arena 1,448 #38 of 205Japanese Human pairwise preferences in the stated text vote pool Arena 1,446 #23 of 152Korean Human pairwise preferences in the stated text vote pool Arena 1,447 #16 of 156non-English Human pairwise preferences in the stated text vote pool Arena 1,439 #37 of 205Polish Human pairwise preferences in the stated text vote pool Arena 1,460 #21 of 142Russian Human pairwise preferences in the stated text vote pool Arena 1,453 #34 of 203Spanish Human pairwise preferences in the stated text vote pool Arena 1,456 #36 of 168overall · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,451 #36 of 188creative writing · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,446 #26 of 187overall · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,451 #36 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 75.6% #26 of 96
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 1,553 #50 of 94
Lech Mazur Short-Story Judges compare stories written to ten required elements; the comparison score can be negative Lech 0.4 #19 of 45
Health & law 5
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 17.8% #30 of 103
Vals LegalBench Legal reasoning and legal-language understanding on LegalBench Vals 82.4% #52 of 89
Vals MedCode Medical coding accuracy on patient encounters Vals 42.5% #42 of 78
Vals MedScribe Accuracy of clinical notes produced from consultation transcripts Vals 80.2% #49 of 80
Vals Legal Research Bench Grounded legal research; a task passes only if all criteria pass Vals 40.9% #24 of 73
Documents 2
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 11.4% #53 of 135
AA-LCR (long context)1.1 Questions that need reading and connecting facts across very long documents (around 100K tokens) AA 80.3% #41 of 317
Legacy · saturated or no longer reported 10
GPQA Diamond Graduate-level biology, chemistry and physics questions that can’t be solved by a quick web search 92.8% #14 of 315+1 versions & slices Epoch Models answer graduate-level science multiple-choice questions Epoch 91.7% #21 of 134
OTIS Mock AIME 20242025 Models solve the stated OTIS mock AIME mathematics problems Epoch 98.6% #20 of 131
SWE-bench VerifiedEpoch Swe Bench Verified: Agents modify real repositories to resolve the stated Verified issues Epoch 77.6% #3 of 24
SWE-rebenchall · March 1, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 41.4% #4 of 6+5 versions & slices Go · March 1, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 48.6% #3 of 6Java · March 1, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 34.3% #4 of 6Python · March 1, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 41.8% #4 of 6Rust · March 1, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 39.2% #4 of 6TypeScript · March 1, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 41.7% #4 of 6
No benchmark matches
CheaperGemini 3.8 Flashhigh+9%-24%*$3.49Sampling parameters are ignored+1
BetterGPT-6.1 Sollow+3%+102%$0.13Remove unsupported parameters+5
CheaperGLM 5.3 FlashmaxOpensame-88%$0.28Defaults to max · Self-hosted+4
Claude Haiku 5.5max-2%-90%$0.21Remove sampling parameters+5
Kimi K3maxOpen+4%+203%$2.00Memory (4-bit): 990 GB → 1,668 GB+4
Muse Spark OpenxhighOpen-2%same$0.84Reasoning off leaves output unchanged · Self-hosted / llama.cpp+1
Qwen3.8 27BxhighOpen-3%-52%$1.15Context: 1M → 262K+5