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 27 #70 of 339
Knowledge 7
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 32.6% #68 of 323+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 40.5% #65 of 323
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA 5.3 #51 of 323
IFBench (instructions) Checks whether the model follows precise formatting and content instructions AA 76% #15 of 248
EQ-Bench3 Judge-rated responses to social and emotionally nuanced scenarios, scored through the source pairwise ranking EQ-Bench 1,548 #10 of 53
SimpleQA Verifiedv1 Accuracy on short factual questions with verified answers, answered without web access 34.9% #48 of 84
Mystery Game Puzzles Models choose the best move in puzzle positions from an undisclosed game Epoch 18% #43 of 65
Reasoning 1
Chess Puzzles Models choose the best next move from a novel chess position Epoch 20% #55 of 69
Math 14
MathArenaAIME 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 95.8% #12 of 24+9 versions & slices HMMT February 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 94.7% #7 of 24Apex Answer accuracy on the named mathematics competition under the recorded effort MathArena 24% #12 of 34Apex Shortlist Answer accuracy on the named mathematics competition under the recorded effort MathArena 77.1% #10 of 30ArXivMath · February 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 43% #9 of 20ArXivMath · January 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 71.7% #5 of 18ArXivMath · March 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 55.8% #7 of 14BrokenArXiv · February 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 11.7% #9 of 15BrokenArXiv · March 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 15.6% #5 of 13USAMO 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 51.2% #5 of 8
FrontierMath Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 39% #8 of 40
FrontierMath Tier4 Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 14.6% #9 of 33
FrontierMath Tier 4 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 25.6% #37 of 55
FrontierMath Tiers 1–3 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 57.2% #39 of 70
Science 3
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 37.5% #57 of 339
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 8% #71 of 326
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 51.5% #46 of 142
Coding 8
Terminal-Bench4.0 Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite 0.5% #65 of 134+1 versions & slices 2.1 Older, easier version of Terminal-Bench; Still widely reported AA 65.9% #42 of 156
ITBench SRE IT incident tasks: diagnose and fix broken systems like a site-reliability engineer AA 31.2% #28 of 37
WebDev Arena Human pairwise preferences in the stated webdev vote pool Arena 1,509 #44 of 97+2 versions & slices October 23, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,509 #45 of 98October 19, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,509 #44 of 97
DeepSWE1 Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition Datacurve 23.9% #9 of 20
Terminal-Bench Hard Agents solve difficult software, system-administration and data-processing tasks in a terminal AA 43.9% #22 of 242
Agents 8
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 13% #75 of 135
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 819 #76 of 142
APEX-Agents Agents complete long-horizon professional-services tasks across realistic applications AA 28.5% #12 of 29
EnterpriseOps-Gym-AA Agents complete stateful enterprise workflows in eight domains; final database state determines success AA 38.5% #24 of 43
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 1,038 #66 of 177
EBR Bench Agents learn to play Earthborne Rangers across repeated playthroughs Epoch 2.4% #23 of 24
Furniture Assembly Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step Epoch 21.7% #28 of 31
τ³-BankingAA Agents retrieve banking knowledge and execute tools to resolve customer-support workflows AA 23.3% #45 of 134
Multimodal 3
MMMU-Pro (vision) College-level questions that require understanding images: charts, diagrams, photos AA 79.4% #24 of 128
Arena Vision Human pairwise preferences in the stated vision vote pool Arena 1,282 #23 of 88
Chartography Expert chart reasoning over professional scientific, medical, financial and engineering charts Surge 12.2% #34 of 37
Writing 20
Arena Textcreative writing · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,434 #40 of 197+16 versions & slices creative writing Human pairwise preferences in the stated text vote pool Arena 1,432 #44 of 195overall Human pairwise preferences in the stated text vote pool Arena 1,461 #38 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,520 #22 of 198English Human pairwise preferences in the stated text vote pool Arena 1,460 #33 of 205French Human pairwise preferences in the stated text vote pool Arena 1,471 #34 of 164German Human pairwise preferences in the stated text vote pool Arena 1,451 #37 of 167instruction following Human pairwise preferences in the stated text vote pool Arena 1,451 #36 of 205Japanese Human pairwise preferences in the stated text vote pool Arena 1,443 #24 of 152Korean Human pairwise preferences in the stated text vote pool Arena 1,427 #27 of 156non-English Human pairwise preferences in the stated text vote pool Arena 1,446 #32 of 205Polish Human pairwise preferences in the stated text vote pool Arena 1,438 #41 of 142Russian Human pairwise preferences in the stated text vote pool Arena 1,446 #40 of 203Spanish Human pairwise preferences in the stated text vote pool Arena 1,463 #27 of 168overall · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,455 #33 of 188creative writing · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,433 #40 of 187overall · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,455 #33 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 78.5% #18 of 96
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 1,721 #31 of 94
Lech Mazur Short-Story Judges compare stories written to ten required elements; the comparison score can be negative Lech 0 #24 of 45
Health & law 5
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 11.1% #52 of 103
Vals LegalBench Legal reasoning and legal-language understanding on LegalBench Vals 84.7% #26 of 89
Vals MedCode Medical coding accuracy on patient encounters Vals 40.1% #52 of 78
Vals MedScribe Accuracy of clinical notes produced from consultation transcripts Vals 78.2% #53 of 80
Vals Legal Research Bench Grounded legal research; a task passes only if all criteria pass Vals 15.9% #55 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 81% #36 of 317
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 13% #46 of 135
Arena Document Human pairwise preferences in the stated document vote pool Arena 1,451 #21 of 33
Legacy · saturated or no longer reported 11
GPQA Diamond Graduate-level biology, chemistry and physics questions that can’t be solved by a quick web search 91.1% #27 of 315+1 versions & slices Epoch Models answer graduate-level science multiple-choice questions Epoch 90.8% #28 of 134
OTIS Mock AIME 20242025 Models solve the stated OTIS mock AIME mathematics problems Epoch 96.1% #30 of 131
SWE-bench VerifiedEpoch Swe Bench Verified: Agents modify real repositories to resolve the stated Verified issues Epoch 76.7% #6 of 24
SWE-rebenchall · March 1, 2026 to May 15, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 46.5% #4 of 5+5 versions & slices Go · March 1, 2026 to May 15, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 0% #1 of 5Java · March 1, 2026 to May 15, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 0% #1 of 5Python · March 1, 2026 to May 15, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 46.5% #4 of 5Rust · March 1, 2026 to May 15, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 0% #1 of 5TypeScript · March 1, 2026 to May 15, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 0% #1 of 5
τ²-benchAA Tests service agents that use tools while coordinating with a simulated user AA 95.9% #12 of 245
No benchmark matches
CheaperGemini 3.6 Flashmedium×2.8-12%*$0.75 / $3.75
BetterGPT-6 Lunamax×2.4-90%$0.068Function calling only with none · Chat Completions+3
OpenQwen3.8 Flashxhigh+43%-87%$0.15 / $0.47max_tokens range · QwenCloud+5
CheaperMuse Glimmer 30BxhighOpen+41%-71%$0.30 / $1.10Context: 256K → 131K
GLM 5.3 FlashmaxOpen+34%-60%$0.28Defaults to max · Self-hosted+4
Grok 4.5high+39%+8%$0.76