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 48.1 #10 of 339
Knowledge 5
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 58.9% #7 of 323+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 61.2% #110 of 323
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA 33.7 #7 of 323
SimpleQA Verifiedv1 Accuracy on short factual questions with verified answers, answered without web access 63.3% #13 of 84
Mystery Game Puzzles Models choose the best move in puzzle positions from an undisclosed game Epoch 59% #6 of 65
Reasoning 6
ARC-AGI-1public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 99% #2 of 64+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 97.5% #8 of 67
ARC-AGI-2public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 93.2% #8 of 66+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 88.3% #10 of 69
ARC-AGI-3semi-private Measures interactive game solving through relative human action efficiency ARC 30.2% #4 of 19
Chess Puzzles Models choose the best next move from a novel chess position Epoch 43% #25 of 69
Math 7
AA Analyst Agent Data-analyst tasks: the model explores data with tools and reports findings AA 53.8% #5 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 73.2% #11 of 55
FrontierMath Tiers 1–3 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 85.6% #12 of 70
MathArenaArXivMath · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 82.6% #9 of 24+2 versions & slices BrokenArXiv · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 90.7% #5 of 24ArXivLean · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 32.6% #2 of 9
Science 5
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 52.8% #9 of 339
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 28.3% #13 of 326
Vals Science Index Scientific problem solving across the index’s component evaluations Vals 47.9% #6 of 19
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 55.4% #27 of 142
Terminal-Bench Science0.1 Agents complete expert-authored research workflows across five scientific domains in a terminal AA 28.6% #10 of 45
Coding 8 Leads on 2
Terminal-Bench4.0 Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite 46% #10 of 134+1 versions & slices 2.1 Older, easier version of Terminal-Bench; Still widely reported AA 87.6% #7 of 156
DeepSWE1.1 Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition 72.8% #3 of 44
WebDev Arena Human pairwise preferences in the stated webdev vote pool Arena 1,659 #11 of 97+2 versions & slices October 23, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,659 #12 of 98October 19, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,659 #12 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 99.4% #1 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 98% #1 of 16
Agents 8 Leads on 1
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 53.6% #33 of 135
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 1,558 #13 of 142
EnterpriseOps-Gym-AA Agents complete stateful enterprise workflows in eight domains; final database state determines success AA 47.5% #5 of 43
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 1,595 #24 of 177
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 48.7% #1 of 21
EBR Bench Agents learn to play Earthborne Rangers across repeated playthroughs Epoch 45.7% #7 of 24
Furniture Assembly Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step Epoch 60.8% #7 of 31
τ³-BankingAA Agents retrieve banking knowledge and execute tools to resolve customer-support workflows AA 44.7% #12 of 134
Multimodal 3
MMMU-Pro (vision) College-level questions that require understanding images: charts, diagrams, photos AA 82.4% #10 of 128
Arena Vision Human pairwise preferences in the stated vision vote pool Arena 1,320 #3 of 88
Chartography Expert chart reasoning over professional scientific, medical, financial and engineering charts Surge 27.3% #17 of 37
Writing 20 Leads on 1
Arena Textcreative writing · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,486 #8 of 197+16 versions & slices creative writing Human pairwise preferences in the stated text vote pool Arena 1,471 #10 of 195overall Human pairwise preferences in the stated text vote pool Arena 1,490 #11 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,567 #4 of 198English Human pairwise preferences in the stated text vote pool Arena 1,503 #7 of 205French Human pairwise preferences in the stated text vote pool Arena 1,518 #4 of 164German Human pairwise preferences in the stated text vote pool Arena 1,503 #3 of 167instruction following Human pairwise preferences in the stated text vote pool Arena 1,514 #5 of 205Japanese Human pairwise preferences in the stated text vote pool Arena 1,518 #3 of 152Korean Human pairwise preferences in the stated text vote pool Arena 1,511 #2 of 156non-English Human pairwise preferences in the stated text vote pool Arena 1,497 #4 of 205Polish Human pairwise preferences in the stated text vote pool Arena 1,521 #3 of 142Russian Human pairwise preferences in the stated text vote pool Arena 1,507 #6 of 203Spanish Human pairwise preferences in the stated text vote pool Arena 1,507 #5 of 168overall · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,503 #4 of 188creative writing · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,487 #8 of 187overall · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,503 #4 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 86.3% #1 of 96
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 2,134 #3 of 94
Lech Mazur Short-Story Judges compare stories written to ten required elements; the comparison score can be negative Lech 3.3 #4 of 45
Health & law 5 Leads on 2
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 11.7% #48 of 103
Vals LegalBench Legal reasoning and legal-language understanding on LegalBench Vals 87% #6 of 89
Vals MedCode Medical coding accuracy on patient encounters Vals 63.6% #1 of 78
Vals MedScribe Accuracy of clinical notes produced from consultation transcripts Vals 91% #4 of 80
Vals Legal Research Bench Grounded legal research; a task passes only if all criteria pass Vals 55.3% #1 of 73
Documents 3 Leads on 1
AA-LCR (long context)1.1 Questions that need reading and connecting facts across very long documents (around 100K tokens) AA 79% #58 of 317
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 19.6% #26 of 135
Arena Document Human pairwise preferences in the stated document vote pool Arena 1,516 #1 of 33
Legacy · saturated or no longer reported 9
GPQA Diamond Graduate-level biology, chemistry and physics questions that can’t be solved by a quick web search 93.7% #7 of 315+1 versions & slices Epoch Models answer graduate-level science multiple-choice questions Epoch 93.9% #11 of 134
OTIS Mock AIME 20242025 Models solve the stated OTIS mock AIME mathematics problems Epoch 98.9% #16 of 131
SWE-rebenchall · May 15, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 63.4% #3 of 6+5 versions & slices Go · May 15, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 73.3% #2 of 6Java · May 15, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 48.6% #5 of 6Python · May 15, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 53% #3 of 6Rust · May 15, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 61.6% #4 of 6TypeScript · May 15, 2026 to July 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 78.3% #1 of 6
No benchmark matches
OpenMiMo v2.6 Prounstatedwins 47%-95%$0.43 / $0.87Pass back reasoning_content+3
CheaperGemini 4 Argonhighsame-55%*$1.62Synchronous output stops at 65,536+2
CheaperKimi K3maxOpenwins 46%-45%$2.00Sampling parameters are fixed+3
GLM 5.3maxOpenwins 45%-44%$2.01No image input+4