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 19.3 #125 of 339
Knowledge 6
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 28.4% #88 of 323+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 52.7% #98 of 323
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA -9.4 #92 of 323
IFBench (instructions) Checks whether the model follows precise formatting and content instructions AA 42.7% #143 of 248
EQ-Bench3 Judge-rated responses to social and emotionally nuanced scenarios, scored through the source pairwise ranking EQ-Bench 1,535 #15 of 53
SimpleQA Verifiedv1 Accuracy on short factual questions with verified answers, answered without web access 29.2% #60 of 84
Reasoning 5
ARC-AGI-1public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 73.8% #43 of 64+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 63.7% #46 of 67
ARC-AGI-2public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 14.7% #42 of 66+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 13.6% #46 of 69
Chess Puzzles Models choose the best next move from a novel chess position Epoch 6% #64 of 69
Math 14
FrontierMath Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 9.3% #26 of 40
FrontierMath Tier4 Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 2.1% #23 of 33
MathArenaAIME 2025 Answer accuracy on the named mathematics competition under the recorded effort MathArena 84.2% #19 of 30+11 versions & slices Apex Answer accuracy on the named mathematics competition under the recorded effort MathArena 1.6% #23 of 34BRUMO 2025 Answer accuracy on the named mathematics competition under the recorded effort MathArena 90.8% #13 of 24CMIMC 2025 Answer accuracy on the named mathematics competition under the recorded effort MathArena 66.3% #19 of 22HMMT February 2025 Answer accuracy on the named mathematics competition under the recorded effort MathArena 67.5% #21 of 30Kangaroo 2025 1-2 Answer accuracy on the named mathematics competition under the recorded effort MathArena 61.5% #15 of 17Kangaroo 2025 11-12 Answer accuracy on the named mathematics competition under the recorded effort MathArena 87.5% #14 of 17Kangaroo 2025 3-4 Answer accuracy on the named mathematics competition under the recorded effort MathArena 62.5% #13 of 17Kangaroo 2025 5-6 Answer accuracy on the named mathematics competition under the recorded effort MathArena 68.3% #11 of 17Kangaroo 2025 7-8 Answer accuracy on the named mathematics competition under the recorded effort MathArena 80% #15 of 16Kangaroo 2025 9-10 Answer accuracy on the named mathematics competition under the recorded effort MathArena 95% #11 of 16SMT 2025 Answer accuracy on the named mathematics competition under the recorded effort MathArena 84% #15 of 24
Science 3
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 7.2% #224 of 339
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 0% #181 of 326
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 45.7% #72 of 142
Coding 6
Terminal-Bench4.0 Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite 0% #77 of 134+1 versions & slices 2.1 Older, easier version of Terminal-Bench; Still widely reported AA 55.8% #59 of 156
WebDev Arena Human pairwise preferences in the stated webdev vote pool Arena 1,394 #73 of 97+2 versions & slices October 23, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,394 #74 of 98October 19, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,394 #73 of 97
Terminal-Bench Hard Agents solve difficult software, system-administration and data-processing tasks in a terminal AA 28.8% #61 of 242
Agents 8
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 14.1% #72 of 135
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 713 #86 of 142
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 907 #80 of 177
MCP-Atlas Public MCPAtlas Public: MCPAtlas Public as reported in this launch evaluation; the post is the authority for the recorded task and grading scope 62% #11 of 16
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 25.3% #16 of 21
EBR Bench Agents learn to play Earthborne Rangers across repeated playthroughs Epoch 2.4% #23 of 24
MCP-Atlasfull 1000 Tool orchestration across MCP servers on the stated task split Scale 59.5% #11 of 16
τ³-BankingAA Agents retrieve banking knowledge and execute tools to resolve customer-support workflows AA 24.5% #43 of 134
Multimodal 2
MMMU-Pro (vision) College-level questions that require understanding images: charts, diagrams, photos AA 65.2% #82 of 128
MMMUvalidation Image-grounded university-level questions across 30 disciplines; validation and test sets remain separate MMMU 77.8% #8 of 41
Writing 19
Arena Textcreative writing · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,442 #29 of 197+16 versions & slices creative writing Human pairwise preferences in the stated text vote pool Arena 1,454 #21 of 195overall Human pairwise preferences in the stated text vote pool Arena 1,456 #43 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,460 #77 of 198English Human pairwise preferences in the stated text vote pool Arena 1,450 #51 of 205French Human pairwise preferences in the stated text vote pool Arena 1,458 #49 of 164German Human pairwise preferences in the stated text vote pool Arena 1,427 #55 of 167instruction following Human pairwise preferences in the stated text vote pool Arena 1,459 #30 of 205Japanese Human pairwise preferences in the stated text vote pool Arena 1,390 #58 of 152Korean Human pairwise preferences in the stated text vote pool Arena 1,401 #43 of 156non-English Human pairwise preferences in the stated text vote pool Arena 1,425 #57 of 205Polish Human pairwise preferences in the stated text vote pool Arena 1,423 #62 of 142Russian Human pairwise preferences in the stated text vote pool Arena 1,437 #46 of 203Spanish Human pairwise preferences in the stated text vote pool Arena 1,458 #31 of 168overall · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,438 #52 of 188creative writing · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,442 #29 of 187overall · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,438 #54 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 71.4% #37 of 96
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 1,677 #35 of 94
Health & law 4
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 10.6% #54 of 103
Vals LegalBench Legal reasoning and legal-language understanding on LegalBench Vals 84.1% #34 of 89
Vals MedCode Medical coding accuracy on patient encounters Vals 44.1% #34 of 78
Vals MedScribe Accuracy of clinical notes produced from consultation transcripts Vals 84.1% #34 of 80
Documents 3
AA-LCR (long context)1.1 Questions that need reading and connecting facts across very long documents (around 100K tokens) AA 54% #171 of 317
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 5.2% #87 of 135
Arena Document Human pairwise preferences in the stated document vote pool Arena 1,450 #22 of 33
Legacy · saturated or no longer reported 24
AIME2025 · AA archived AA evaluates the 30 integer-answer problems from the 2025 AIME mathematics competition AA 37% #99 of 140
LiveCodeBenchAA archived AA evaluates solutions to newly collected competitive-programming problems AA 59% #75 of 153
GPQA Diamond Graduate-level biology, chemistry and physics questions that can’t be solved by a quick web search 72.7% #170 of 315+1 versions & slices Epoch Models answer graduate-level science multiple-choice questions Epoch 73.7% #77 of 134
BFCLv4 Tests function calling, multi-turn interactions, search and memory in the v4 suite Berkeley 73.2% #2 of 39+1 versions & slices v4 · prompt mode Tests function calling, multi-turn interactions, search and memory in the v4 suite Berkeley 24.9% #15 of 16
SWE-bench Pro (public) Agents resolve long-horizon professional repository tasks; public and private splits are separate 43.6% #7 of 19
τ²-benchairline · 0.1.3 τ-bench official · 2-airline-0.1.3: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools 70% #1 of 8+8 versions & slices airline · 1.0.1 τ-bench official · 2-airline-1.0.1: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools Sierra 72% #6 of 6overall · 0.1.3 τ-bench official · 2-overall-0.1.3: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools 84.7% #1 of 7overall · 1.0.1 τ-bench official · 2-overall-1.0.1: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools Sierra 76.4% #6 of 6retail · 0.1.3 τ-bench official · 2-retail-0.1.3: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools 86.2% #1 of 8retail · 1.0.1 τ-bench official · 2-retail-1.0.1: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools Sierra 72.4% #6 of 6telecom · 0.1.3 τ-bench official · 2-telecom-0.1.3: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools 98% #2 of 7telecom · 1.0.1 τ-bench official · 2-telecom-1.0.1: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools Sierra 84.9% #6 of 6AA Tests service agents that use tools while coordinating with a simulated user AA 70.5% #94 of 245
OTIS Mock AIME 20242025 Models solve the stated OTIS mock AIME mathematics problems Epoch 35.6% #101 of 131
SWE-bench VerifiedEpoch Swe Bench Verified: Agents modify real repositories to resolve the stated Verified issues Epoch 71.3% #18 of 24
SWE-rebenchall · September 1, 2025 to March 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 52.4% #1 of 1+5 versions & slices Go · September 1, 2025 to March 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 0% #1 of 1Java · September 1, 2025 to March 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 0% #1 of 1Python · September 1, 2025 to March 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 52.4% #1 of 1Rust · September 1, 2025 to March 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 0% #1 of 1TypeScript · September 1, 2025 to March 1, 2026 Repository issue resolution over the exact fresh pull-request window, under the source agent version Nebius 0% #1 of 1
No benchmark matches
CheaperGemini 3.1 Pro Previewunstatedwins 58%-25%$2 / $12Cannot use both thinking controls
BetterGPT-5.5xhighwins 61%+88%$2.10Prompts with >272K input tokens
OpenKimi K3unstatedwins 59%same$3 / $15Sampling parameters are fixed+3
CheaperDeepSeek V4 Pro 0813maxOpenwins 54%-67%$1.32 / $3.96No image input+5
GLM 5.2unstatedOpenwins 53%-64%$1.40 / $4.40No image input