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 39.2 #34 of 339
Knowledge 6
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 57.8% #8 of 323+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 90.8% #257 of 323
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA 19.4 #28 of 323
IFBench (instructions) Checks whether the model follows precise formatting and content instructions AA 69.6% #52 of 248
SimpleQA Verifiedv1 Accuracy on short factual questions with verified answers, answered without web access 69.7% #9 of 84
Mystery Game Puzzles Models choose the best move in puzzle positions from an undisclosed game Epoch 58% #7 of 65
Reasoning 6
ARC-AGI-1public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 93.9% #29 of 64+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 92.5% #21 of 67
ARC-AGI-2public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 70.7% #19 of 66+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 67.1% #23 of 69
ARC-AGI-3semi-private Measures interactive game solving through relative human action efficiency ARC 1.1% #11 of 19
Chess Puzzles Models choose the best next move from a novel chess position Epoch 59% #6 of 69
Math 6
AA Analyst Agent Data-analyst tasks: the model explores data with tools and reports findings AA 47.5% #12 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 82.9% #7 of 55
FrontierMath Tiers 1–3 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 89.1% #6 of 70
MathArenaArXivMath · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 88.5% #5 of 24+1 versions & slices BrokenArXiv · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 67.3% #8 of 24
Science 5
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 42.2% #39 of 339
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 22.9% #24 of 326
Vals Science Index Scientific problem solving across the index’s component evaluations Vals 41.5% #8 of 19
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 57.4% #18 of 142
Terminal-Bench Science0.1 Agents complete expert-authored research workflows across five scientific domains in a terminal AA 22.4% #12 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 14.6% #32 of 134+1 versions & slices 2.1 Older, easier version of Terminal-Bench; Still widely reported AA 86.1% #8 of 156
DeepSWE1.1 Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition 61.1% #19 of 44
ITBench SRE IT incident tasks: diagnose and fix broken systems like a site-reliability engineer AA 56.2% #1 of 37
SWE-bench Pro (v2 full) Agents resolve repository issues under the network-locked v2 protocol; full and 51-task Hard results are separate Scale 95.5% #10 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 82.4% #10 of 16
MirrorCode Agents reimplement software projects by observing their behavior Epoch 20% #7 of 10
Terminal-Bench Hard Agents solve difficult software, system-administration and data-processing tasks in a terminal AA 62.9% #1 of 242
Agents 8
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 51.3% #35 of 135
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 1,240 #43 of 142
EnterpriseOps-Gym-AA Agents complete stateful enterprise workflows in eight domains; final database state determines success AA 42.9% #15 of 43
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 1,422 #40 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 46.9% #3 of 21
EBR Bench Agents learn to play Earthborne Rangers across repeated playthroughs Epoch 44.8% #8 of 24
Furniture Assembly Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step Epoch 56.7% #9 of 31
τ³-BankingAA Agents retrieve banking knowledge and execute tools to resolve customer-support workflows AA 36.5% #22 of 134
Multimodal 3
MMMU-Pro (vision) College-level questions that require understanding images: charts, diagrams, photos AA 81.4% #12 of 128
Arena Vision Human pairwise preferences in the stated vision vote pool Arena 1,281 #24 of 88
Chartography Expert chart reasoning over professional scientific, medical, financial and engineering charts Surge 45% #6 of 37
Writing 20
Arena Textcreative writing · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,450 #24 of 197+16 versions & slices creative writing Human pairwise preferences in the stated text vote pool Arena 1,469 #12 of 195overall Human pairwise preferences in the stated text vote pool Arena 1,485 #16 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,526 #20 of 198English Human pairwise preferences in the stated text vote pool Arena 1,458 #37 of 205French Human pairwise preferences in the stated text vote pool Arena 1,472 #33 of 164German Human pairwise preferences in the stated text vote pool Arena 1,475 #16 of 167instruction following Human pairwise preferences in the stated text vote pool Arena 1,483 #13 of 205Japanese Human pairwise preferences in the stated text vote pool Arena 1,473 #14 of 152Korean Human pairwise preferences in the stated text vote pool Arena 1,442 #18 of 156non-English Human pairwise preferences in the stated text vote pool Arena 1,451 #28 of 205Polish Human pairwise preferences in the stated text vote pool Arena 1,458 #24 of 142Russian Human pairwise preferences in the stated text vote pool Arena 1,472 #22 of 203Spanish Human pairwise preferences in the stated text vote pool Arena 1,442 #54 of 168overall · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,457 #32 of 188creative writing · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,449 #24 of 187overall · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,457 #32 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 81.7% #12 of 96
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 1,974 #10 of 94
Lech Mazur Short-Story Judges compare stories written to ten required elements; the comparison score can be negative Lech 2.4 #8 of 45
Health & law 5
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 26.1% #13 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 44% #35 of 78
Vals MedScribe Accuracy of clinical notes produced from consultation transcripts Vals 85.2% #28 of 80
Vals Legal Research Bench Grounded legal research; a task passes only if all criteria pass Vals 48.1% #8 of 73
Documents 3
AA-LCR (long context)1.1 Questions that need reading and connecting facts across very long documents (around 100K tokens) AA 80.3% #40 of 317
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 26.2% #9 of 135
Arena Document Human pairwise preferences in the stated document vote pool Arena 1,483 #7 of 33
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 92.6% #16 of 315+1 versions & slices Epoch Models answer graduate-level science multiple-choice questions Epoch 93.5% #12 of 134
OTIS Mock AIME 20242025 Models solve the stated OTIS mock AIME mathematics problems Epoch 100% #1 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 62.3% #4 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 70.5% #3 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 57.1% #2 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 48% #5 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 68% #1 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 65.8% #4 of 6
τ²-benchAA Tests service agents that use tools while coordinating with a simulated user AA 81% #76 of 245
No benchmark matches
OpenDeepSeek V4.1 Flash 1017max+16%-48%$0.26Context: 1.05M → 1M+5
CheaperGemini 4 Argonmedium+30%-50%*$2 / $10Context: 1.05M → 1M+3
Claude Opus 5.5low+5%+9%$0.55Context: 1.05M → 1M+6
GLM 5.3 FlashmaxOpen-2%-44%$0.28Context: 1.05M → 1M+5
Kimi K3maxOpen-10%+296%$2.00Context: 1.05M → 1M+4
Muse Spark OpenxhighOpen-14%+67%$0.84Context: 1.05M → 1M+2