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 15.4 #145 of 339
Knowledge 5
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 14.4% #245 of 323+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 25.7% #22 of 323
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA -7.6 #88 of 323
IFBench (instructions) Checks whether the model follows precise formatting and content instructions AA 42% #143 of 248
SimpleQA Verifiedv1 Accuracy on short factual questions with verified answers, answered without web access 13.2% #73 of 84
Reasoning 4
ARC-AGI-1public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 62.9% #49 of 64+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 47.7% #54 of 67
ARC-AGI-2public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 5.1% #50 of 66+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 4% #57 of 69
Math 3
AA Analyst Agent Data-analyst tasks: the model explores data with tools and reports findings AA 15% #35 of 43
FrontierMath Models solve expert-written advanced mathematics problems on the named tier and edition Epoch 4.1% #33 of 40
MATH Level5 Models solve competition mathematics problems from the hardest MATH difficulty level Epoch 86.9% #8 of 35
Science 3
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 4.2% #304 of 339
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 0% #182 of 326
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 42.2% #87 of 142
Coding 9
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 44.2% #78 of 156
ITBench SRE IT incident tasks: diagnose and fix broken systems like a site-reliability engineer AA 27.3% #31 of 37
WebDev Arena Human pairwise preferences in the stated webdev vote pool Arena 1,330 #84 of 97+2 versions & slices October 23, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,330 #85 of 98October 19, 2026 Human pairwise preferences in the stated webdev vote pool Arena 1,330 #84 of 97
SWE-bench Pro (v2 hard) Agents resolve repository issues under the network-locked v2 protocol; full and 51-task Hard results are separate Scale 25.5% #16 of 16
DeepSWE1 Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition Datacurve 0.2% #19 of 20
Terminal-Bench Hard Agents solve difficult software, system-administration and data-processing tasks in a terminal AA 27.3% #67 of 242
Agents 7
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 3.2% #92 of 135
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 616 #92 of 142
EnterpriseOps-Gym-AA Agents complete stateful enterprise workflows in eight domains; final database state determines success AA 29.4% #35 of 43
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 732 #92 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 41.2% #16 of 16
MCP-Atlasfull 1000 Tool orchestration across MCP servers on the stated task split Scale 40.2% #16 of 16
τ³-BankingAA Agents retrieve banking knowledge and execute tools to resolve customer-support workflows AA 9.3% #95 of 134
Multimodal 1
MMMU-Pro (vision) College-level questions that require understanding images: charts, diagrams, photos AA 55.1% #108 of 128
Writing 18
Arena Textcreative writing · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,371 #85 of 197+16 versions & slices creative writing Human pairwise preferences in the stated text vote pool Arena 1,388 #80 of 195overall Human pairwise preferences in the stated text vote pool Arena 1,414 #88 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,418 #106 of 198English Human pairwise preferences in the stated text vote pool Arena 1,414 #93 of 205French Human pairwise preferences in the stated text vote pool Arena 1,410 #94 of 164German Human pairwise preferences in the stated text vote pool Arena 1,374 #89 of 167instruction following Human pairwise preferences in the stated text vote pool Arena 1,414 #71 of 205Japanese Human pairwise preferences in the stated text vote pool Arena 1,338 #83 of 152Korean Human pairwise preferences in the stated text vote pool Arena 1,346 #87 of 156non-English Human pairwise preferences in the stated text vote pool Arena 1,377 #104 of 205Polish Human pairwise preferences in the stated text vote pool Arena 1,343 #103 of 142Russian Human pairwise preferences in the stated text vote pool Arena 1,381 #100 of 203Spanish Human pairwise preferences in the stated text vote pool Arena 1,422 #74 of 168overall · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,397 #94 of 188creative writing · October 15, 2026 Human pairwise preferences in the stated text vote pool Arena 1,371 #82 of 187overall · October 22, 2026 Human pairwise preferences in the stated text vote pool Arena 1,397 #99 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 65% #48 of 96
Health & law 5
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 4.4% #83 of 103
Vals LegalBench Legal reasoning and legal-language understanding on LegalBench Vals 81.2% #60 of 89
Vals MedCode Medical coding accuracy on patient encounters Vals 32.7% #66 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 10.6% #64 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 49.7% #181 of 317
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 3.8% #96 of 135
Arena Document Human pairwise preferences in the stated document vote pool Arena 1,420 #29 of 33
Legacy · saturated or no longer reported 9
AIME2025 · AA archived AA evaluates the 30 integer-answer problems from the 2025 AIME mathematics competition AA 39% #97 of 140
LiveCodeBenchAA archived AA evaluates solutions to newly collected competitive-programming problems AA 51.1% #89 of 153
GPQA Diamond Graduate-level biology, chemistry and physics questions that can’t be solved by a quick web search 64.6% #213 of 315+1 versions & slices Epoch Models answer graduate-level science multiple-choice questions Epoch 60.5% #95 of 134
BFCLv4 Tests function calling, multi-turn interactions, search and memory in the v4 suite Berkeley 68.7% #3 of 39+1 versions & slices v4 · prompt mode Tests function calling, multi-turn interactions, search and memory in the v4 suite Berkeley 25.3% #14 of 16
SWE-bench Pro (public) Agents resolve long-horizon professional repository tasks; public and private splits are separate 39.5% #10 of 19
OTIS Mock AIME 20242025 Models solve the stated OTIS mock AIME mathematics problems Epoch 35.8% #100 of 131
τ²-benchAA Tests service agents that use tools while coordinating with a simulated user AA 32.5% #144 of 245
No benchmark matches
CheaperGemini 3 Flash Previewthinking×2.5-44%$0.50 / $3
BetterGPT OSS 20BhighOpen×2.3-98%$0.011Context: 200K → 131K+1
OpenDeepSeek V3.2 Specialethinking×2.5——Context: 200K → 128K+1
CheaperGLM 4.7thinkingOpen×2.4-50%$0.60 / $2.20No image input
CheaperKimi K2 ThinkingthinkingOpen×2.4-57%$0.47 / $2No image input
Nova 2.0 Litehigh×2.4-57%$0.30 / $2.50No strict JSON
Nemotron 3 Nano 30B A3BthinkingOpen×2.3-95%$0.017Context: 200K → 131K+1
Qwen3 30B A3B 2507 (Reasoning)thinkingOpen+44%—$0.066No image input