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 24.1 #89 of 337
Knowledge 4
AA OmniscienceAccuracy Measures correct answers on the AA Omniscience factual knowledge test AA 37.5% #55 of 321+1 versions & slices Hallucination Rate Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test AA 90.2% #252 of 321
AA Omniscience Index Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty AA -18.9 #121 of 321
IFBench (instructions) Checks whether the model follows precise formatting and content instructions AA 73.3% #31 of 248
Reasoning 4
ARC-AGI-1public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 75.1% #39 of 61+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 63.7% #43 of 64
ARC-AGI-2public evaluation Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 17.8% #38 of 63+1 versions & slices semi-private Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit ARC 18.9% #40 of 66
Math 1
AA Analyst Agent Data-analyst tasks: the model explores data with tools and reports findings AA 15% #35 of 43
Science 3
Humanity's Last Exam Very hard expert questions across many fields, written to stump today’s models AA 28.1% #92 of 337
CritPt (physics) Research-level physics problems that need many careful reasoning steps AA 10% #59 of 324
SciCode Writing code for real scientific computing problems (physics, biology, math) AA 52.1% #40 of 140
Coding 7
Terminal-Bench4.0 Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite 2% #49 of 132+1 versions & slices 2.1 Older, easier version of Terminal-Bench; Still widely reported AA 59.2% #56 of 156
ITBench SRE IT incident tasks: diagnose and fix broken systems like a site-reliability engineer AA 35.2% #24 of 37
WebDev Arena Human pairwise preferences in the stated webdev vote pool Arena 1,397 #71 of 97+1 versions & slices full cohort Human pairwise preferences in the stated webdev vote pool Arena 1,397 #71 of 97
DeepSWE1 Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition Datacurve 24.3% #8 of 20
Terminal-Bench Hard Agents solve difficult software, system-administration and data-processing tasks in a terminal AA 52.3% #11 of 242
Agents 8
AutomationBench-AA Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation AA 26.2% #51 of 133
AA-Briefcase1.1 · Elo Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating AA 719 #82 of 140
APEX-Agents Agents complete long-horizon professional-services tasks across realistic applications AA 28.2% #13 of 29
EnterpriseOps-Gym-AA Agents complete stateful enterprise workflows in eight domains; final database state determines success AA 34.6% #29 of 43
GDPval-AA (work tasks)2.1 Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work AA 25.6 #76 of 175
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 61% #12 of 16
MCP-Atlasfull 1000 Tool orchestration across MCP servers on the stated task split Scale 56.7% #13 of 16
τ³-BankingAA Agents retrieve banking knowledge and execute tools to resolve customer-support workflows AA 25.6% #41 of 134
Multimodal 2
MMMU-Pro (vision) College-level questions that require understanding images: charts, diagrams, photos AA 73.3% #62 of 127
Arena Vision Human pairwise preferences in the stated vision vote pool Arena 1,245 #48 of 88
Writing 17
Arena Textcreative writing Human pairwise preferences in the stated text vote pool Arena 1,402 #67 of 195+14 versions & slices overall Human pairwise preferences in the stated text vote pool Arena 1,447 #54 of 196Chinese Human pairwise preferences in the stated text vote pool Arena 1,446 #84 of 181English Human pairwise preferences in the stated text vote pool Arena 1,412 #91 of 188French Human pairwise preferences in the stated text vote pool Arena 1,440 #69 of 149German Human pairwise preferences in the stated text vote pool Arena 1,409 #66 of 154instruction following Human pairwise preferences in the stated text vote pool Arena 1,405 #75 of 188Japanese Human pairwise preferences in the stated text vote pool Arena 1,375 #62 of 138Korean Human pairwise preferences in the stated text vote pool Arena 1,371 #64 of 144non-English Human pairwise preferences in the stated text vote pool Arena 1,405 #74 of 188Polish Human pairwise preferences in the stated text vote pool Arena 1,415 #59 of 133Russian Human pairwise preferences in the stated text vote pool Arena 1,417 #66 of 186Spanish Human pairwise preferences in the stated text vote pool Arena 1,404 #82 of 155overall · full cohort Human pairwise preferences in the stated text vote pool Arena 1,411 #82 of 188creative writing · full cohort Human pairwise preferences in the stated text vote pool Arena 1,370 #83 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 73% #32 of 95
EQ-Bench Creative Writingv3 Creative Writing v3 · Elo: Judge-rated creative writing under the v3 prompt set and pairwise ranking EQ-Bench 1,665 #37 of 91
Health & law 2
MLCR-AA Models synthesize long medical records; the headline credits concise answers judged both accurate and complete AA 3.9% #51 of 64
Vals Legal Research Bench Grounded legal research; a task passes only if all criteria pass Vals 12.5% #59 of 69
Documents 2
AA-LCR (long context)1.1 Questions that need reading and connecting facts across very long documents (around 100K tokens) AA 77% #80 of 275
GDP.pdfAA all-pass Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes 13.4% #43 of 133
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 87.5% #52 of 315
τ²-benchAA Tests service agents that use tools while coordinating with a simulated user AA 83.3% #71 of 245
No benchmark matches
OpenKimi K3max+64%+347%$2.00Sampling parameters are fixed+3
Claude Haiku 5.5high+28%-82%$0.079Remove sampling parameters+5
MiMo v2.6 ProthinkingOpen+43%-70%$0.13Pass back reasoning_content+3
Muse Spark OpenxhighOpen+39%+89%$0.84Reasoning off leaves output unchanged · Self-hosted / llama.cpp+1
DeepSeek V4.1 Flash 1017maxOpen+28%-41%$0.26Pass back reasoning_content · Chat Completions+4
Qwen3.8 27BxhighOpen+22%+125%$1.01Context: 400K → 262K+5