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 53.4 #4 of 337
Knowledge 5 Leads on 1
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 67.2% #1 of 321+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 72.6% #136 of 321
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA 43.5 #3 of 321
SimpleQA Verifiedv1 Accuracy on short factual questions with verified answers, answered without web access 70.8% #7 of 83
Mystery Game Puzzles Models choose the best move in puzzle positions from an undisclosed game Epoch 58% #7 of 64
Reasoning 5 Leads on 1
ARC-AGI-1public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 99% #2 of 61+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 64
ARC-AGI-2public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 98.3% #1 of 63+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 90% #7 of 66
Chess Puzzles Models choose the best next move from a novel chess position Epoch 48% #15 of 68
Math 11 Leads on 1
AA Analyst Agent Data-analyst tasks: the model explores data with tools and reports findings AA 57.5% #1 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 87.8% #6 of 54
FrontierMath Tiers 1–3 (v2) Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 90.2% #4 of 69
MathArenaArXivMath · April 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 77.5% #2 of 17+6 versions & slices ArXivMath · August 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 87.7% #4 of 12ArXivMath · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 91% #3 of 24ArXivMath · May 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 87.5% #4 of 19BrokenArXiv · August 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 79.8% #3 of 12BrokenArXiv · June 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 86.6% #6 of 24BrokenArXiv · May 2026 Answer accuracy on the named mathematics competition under the recorded effort MathArena 78% #5 of 18
Science 4
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 59.1% #2 of 337
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 29.7% #11 of 324
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 63.1% #2 of 140
Terminal-Bench Science0.1 Agents complete expert-authored research workflows across five scientific domains in a terminal AA 43.3% #6 of 33
Coding 9
Terminal-Bench4.0 Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite 52% #7 of 132+1 versions & slices 2.1 Older, easier version of Terminal-Bench; Still widely reported AA 91.4% #1 of 156
FrontierCode1.1 Agents implement changes in real repositories, graded against maintainer-defined requirements AA 49% #3 of 28
ITBench SRE IT incident tasks: diagnose and fix broken systems like a site-reliability engineer AA 49.5% #6 of 37
WebDev Arena Human pairwise preferences in the stated webdev vote pool Arena 1,743 #5 of 97+1 versions & slices full cohort Human pairwise preferences in the stated webdev vote pool Arena 1,743 #5 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 99.1% #3 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 92.2% #2 of 16
MirrorCode Agents reimplement software projects by observing their behavior Epoch 73.3% #2 of 9
Agents 9
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 59.4% #21 of 133
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 1,672 #4 of 140
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 62.7 #5 of 175
EBR Bench Agents learn to play Earthborne Rangers across repeated playthroughs Epoch 57.1% #4 of 24
Furniture Assembly Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step Epoch 70% #6 of 31
SpireClimb Agents play Slay the Spire under the named game and difficulty setting Epoch 2% #5 of 7
SpireClimb StS1 A20 Heart Agents play Slay the Spire under the named game and difficulty setting Epoch 0% #4 of 7
SpireClimb StS2 A10 Agents play Slay the Spire under the named game and difficulty setting Epoch 4% #5 of 7
τ³-BankingAA Agents retrieve banking knowledge and execute tools to resolve customer-support workflows AA 47.2% #7 of 134
Multimodal 2
Chartography Expert chart reasoning over professional scientific, medical, financial and engineering charts Surge 46.2% #5 of 37
Arena Vision Human pairwise preferences in the stated vision vote pool Arena 1,319 #4 of 88
Writing 18 Leads on 1
Arena Textcreative writing Human pairwise preferences in the stated text vote pool Arena 1,485 #7 of 195+14 versions & slices overall Human pairwise preferences in the stated text vote pool Arena 1,502 #6 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,586 #2 of 181English Human pairwise preferences in the stated text vote pool Arena 1,505 #6 of 188French Human pairwise preferences in the stated text vote pool Arena 1,527 #1 of 149German Human pairwise preferences in the stated text vote pool Arena 1,502 #4 of 154instruction following Human pairwise preferences in the stated text vote pool Arena 1,517 #5 of 188Japanese Human pairwise preferences in the stated text vote pool Arena 1,560 #1 of 138Korean Human pairwise preferences in the stated text vote pool Arena 1,521 #1 of 144non-English Human pairwise preferences in the stated text vote pool Arena 1,507 #2 of 188Polish Human pairwise preferences in the stated text vote pool Arena 1,524 #1 of 133Russian Human pairwise preferences in the stated text vote pool Arena 1,519 #4 of 186Spanish Human pairwise preferences in the stated text vote pool Arena 1,521 #1 of 155overall · full cohort Human pairwise preferences in the stated text vote pool Arena 1,512 #3 of 188creative writing · full cohort Human pairwise preferences in the stated text vote pool Arena 1,508 #3 of 187
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 85.3% #2 of 95
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 2,162 #2 of 91
Lech Mazur Short-Story Judges compare stories written to ten required elements; the comparison score can be negative Lech 3.8 #1 of 44
Health & law 6 Leads on 1
Harvey LAB-AA1.1 Legal agents work from case documents; every rubric criterion must pass and material hallucinations receive no credit AA 6.4% #7 of 23
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 71.1% #2 of 64
Vals LegalBench Legal reasoning and legal-language understanding on LegalBench Vals 88.5% #2 of 89
Vals MedCode Medical coding accuracy on patient encounters Vals 53.5% #7 of 78
Vals MedScribe Accuracy of clinical notes produced from consultation transcripts Vals 91.3% #2 of 80
Vals Legal Research Bench Grounded legal research; a task passes only if all criteria pass Vals 55.3% #1 of 69
Documents 3
AA-LCR (long context)1.1 Questions that need reading and connecting facts across very long documents (around 100K tokens) AA 85.3% #4 of 275
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 26.2% #8 of 133
Arena Document Human pairwise preferences in the stated document vote pool Arena 1,513 #2 of 33
Legacy · saturated or no longer reported 2
GPQA Diamond Graduate-level biology, chemistry and physics questions that can’t be solved by a quick web search 93.7% #7 of 315
OTIS Mock AIME 20242025 Models solve the stated OTIS mock AIME mathematics problems Epoch 100% #1 of 130
No benchmark matches
CheaperGPT-6.1 Solmax+4%-91%$0.72Remove unsupported parameters+5
OpenKimi K3max-20%-74%$2.00Sampling parameters are fixed+3
GLM 5.3maxOpen-24%-74%$2.01No image input+4
DeepSeek V4 Pro 0813maxOpen-28%-91%$0.67No image input+5