Leaderboards

Every leaderboard on ModelScale

Rank the catalog as a whole, by one capability axis, or by a single benchmark. Every table is built from values the sources published; a model a source never scored is left out of that ranking rather than placed at the bottom of it.

Featured

Both rank the whole catalog. The first uses the published overall score; the second uses weights you choose.

Capability leaderboards

One page per capability axis, ranked by that axis's score rather than the overall one.

Agentic leaderboardHow well a model plans multi-step work, calls tools, and carries a task to completion without a human steering every turn.Coding leaderboardHow well a model writes, debugs and reasons about code across languages and existing codebases.Knowledge leaderboardBreadth and accuracy of factual recall, from general trivia to specialist domains, without external tools or retrieval.Reasoning leaderboardMulti-step logical and analytical problem-solving: the kind of task where the path to the answer matters as much as the answer itself.

Benchmark leaderboards

730 benchmarks in 10 categories, each with its own cross-model ranking.

Agentic107

AA Agentic Index leaderboardA display-only Artificial Analysis agentic index.AA AutomationBench leaderboardAn independently evaluated automation benchmark from Artificial Analysis.AA Briefcase leaderboardAn independently evaluated professional-work benchmark reported as Elo.AA EnterpriseOps-Gym leaderboardAn independently evaluated enterprise-operations benchmark from Artificial Analysis.AA Harvey LAB v1.0 leaderboardArchived v1.0 legal-agent results from Artificial Analysis, kept separate from v1.1.AA ITBench leaderboardAn independently evaluated IT-operations benchmark from Artificial Analysis.AA Tau3 Banking leaderboardAn independently evaluated Tau3 banking benchmark from Artificial Analysis.AA Terminal-Bench 4.0 leaderboardAn independently evaluated Terminal-Bench v4.0 result from Artificial Analysis, one of the ten components of its Intelligence Index v4.3.AA-AnalystAgent leaderboardArtificial Analysis' data-analysis benchmark, testing agents on spreadsheet and document work to answer the quantitative questions a business or data analyst faces day to day.Agents' Last Exam leaderboardAn agent benchmark reported in DeepSeek's V4 Flash 0731 launch comparison.AI4AI-Bench leaderboardTests whether coding agents can improve the training algorithm inside an existing AI research codebase, then survive a sealed training run and held-out evaluation.AndroidWorld leaderboardA mobile GUI agent benchmark for completing Android app workflows and on-device tasks.APEX-Agents leaderboardA professional-services agent benchmark covering long-horizon knowledge-work tasks.APEX-Agents-AA leaderboardArtificial Analysis' implementation of the APEX-Agents benchmark for long-horizon professional-services agent tasks.ApprenticeBench leaderboardTests whether a computer-use agent can learn a real accounts-payable job on the job, processing 100 vendor bills in a company ERP system with only the handbook, historical records, and mentor feedback a new hire would get.AutoCAD-Bench leaderboardA Markov Studios computer-use benchmark that asks agents to produce 2D drawings and 3D models in AutoCAD.AutomationBench leaderboardAn agent benchmark for completing automation workflows in reproducible task environments.AutomationBench (Zapier 1.0.6) leaderboardAn agent benchmark for completing business workflows across simulated applications and API endpoints.BankerToolBench leaderboardA display-only provider benchmark for finance-oriented tool-use and agent workflows.BFCL v3 leaderboardA function-calling benchmark for tool selection, schema adherence, and argument correctness, covering single-turn, parallel, irrelevance and multi-turn subsets.BFCL v4 leaderboardA function-calling benchmark for tool selection, schema adherence, and argument correctness.BrowseComp leaderboardBrowseComp tests whether a browsing agent can find difficult, fact-seeking answers across the web.BrowseComp (10-agent, prerelease) leaderboardBrowseComp accuracy from ten collaborating Opus 5 agents on a pre-release model and unreleased effort configuration.BrowseComp-VL leaderboardA vision-language browsing benchmark for multimodal web research and tool-use workflows.BrowseComp+ leaderboardEvaluates research agents against a fixed document corpus, separately from live-web BrowseComp.Claw-Eval leaderboardA transparent real-world autonomous-agent benchmark with 300 human-verified tasks, 2,159 rubric items, and Pass^3 scoring across general, multi-turn, and native multimodal agent tasks.CoWorkBench leaderboardQwen's internal benchmark for long-horizon professional work across computer science, finance, law, medicine, and other productivity domains.CTI-REALM leaderboardA cybersecurity benchmark that measures whether an agent can turn raw threat-intelligence reports into working detection rules.CWE-Bench leaderboardAn external benchmark for evaluating whether coding agents can produce correct patches for real-world software vulnerabilities.CWE-bench v1 leaderboardDefensive vulnerability patching on 120 held-out tasks spanning 73 weakness types.Cybench leaderboardA cybersecurity benchmark of professional Capture the Flag tasks for measuring autonomous cyber agent capability and risk.CyberGym leaderboardA cybersecurity task benchmark for evaluating defensive cyber workflows and vulnerability-oriented agent performance.DECK-Bench leaderboardMoonshot AI's internal benchmark for presentation and deck-production workflows.DeepPlanning leaderboardA long-horizon planning benchmark that tests whether agents can optimize under explicit time, budget, and feasibility constraints.DeepSearchQA leaderboardAn agentic browsing benchmark where models search the web, gather evidence, and answer list-style questions using browser tools.Design Arena Agentic Web Dev leaderboardA display-only Elo rating from blinded comparisons of multi-file web applications built by coding agents.DRACO leaderboardAgentic data-analysis tasks scored against per-task rubrics at a 980K-token budget.EdgeBench leaderboardA ByteDance Seed benchmark of 134 real-world, day-scale tasks that measures how autonomous agents learn from environment feedback over 12+ hour interaction horizons, spanning scientific and ML, systems and software engineering, optimization, knowledge, formal, and game domains.EQ-Bench 4 leaderboardA multi-turn benchmark of emotional and social intelligence using synthetic personas and pairwise LLM judging.ExploitGym leaderboardA controlled benchmark for evaluating whether AI agents can extend vulnerability-triggering inputs into working exploits.ExplorationBench AlienCode leaderboardTests whether a system can discover a programming language's hidden rules through experiments and apply them to unseen tasks.ExplorationBench AlienLogic leaderboardTests whether a system can discover altered logic rules and use them to prove or correctly decline unseen theorems.Finance Agent v2 leaderboardVals AI benchmark for realistic financial analyst agent tasks across qualitative analysis, quantitative analysis, market work, comparables, precedents, earnings, disclosure, and modeling.GAIA leaderboardGAIA evaluates AI models on real-world tasks that are conceptually simple for humans but require multi-step reasoning, web browsing, tool use, and multimodal understanding for AI.GDP.pdf leaderboardArtificial Analysis' document-output benchmark: professional tasks whose deliverable is a produced PDF, one of the ten components of its Intelligence Index v4.3.GDPval rubrics leaderboardA display-only provider-table GDPval rubric score for economically valuable work tasks.GDPval-AA leaderboardAn agentic real-world work-task evaluation reported as an Elo score in DeepSeek-V4 thinking-mode evaluations.GDPval-AA leaderboardAn Artificial Analysis normalized score for economically valuable tasks.Gert Labs leaderboardA game-environment benchmark that evaluates AI models in novel games covering strategic planning, resource management, spatial reasoning, cooperation, and theory of mind.Harvey LAB-AA v1.1 leaderboardLegal-agent task success requiring every rubric criterion to pass and no material hallucination, averaged across three judges.Harvey LAB-AA v1.1 Criterion Pass leaderboardThe share of legal-deliverable rubric criteria passed, pooled across three judges before the hallucination gate.Hermes Bench leaderboardTests agents on 150 tasks across 25 categories, including Hermes skills, research, diagrams and art, memory, tool use and safety.Hermes Index leaderboardAn index of agentic capability from Nous Research.HLE w/ tools leaderboardTool-augmented Humanity's Last Exam scores reported in DeepSeek-V4 thinking-mode evaluations.InferenceBench leaderboardA benchmark for open-ended LLM inference optimization by AI agents.JobBench leaderboardAn occupational agent benchmark for professional workflows that workers say they most want delegated to AI.Kimi Claw 24/7 leaderboardA Moonshot AI internal long-horizon agent benchmark for persistent professional coworking tasks.LAB all-pass (Anthropic harness) leaderboardStrict task success requiring every expert-written legal-work rubric criterion to pass.LAB all-pass (Harvey held-out) leaderboardHarvey AI's strict held-out task success rate requiring every rubric criterion to pass.LAB criterion-pass (Anthropic harness) leaderboardMean fraction of expert-written rubric criteria passed across legal-agent tasks.LAB criterion-pass (Harvey held-out) leaderboardHarvey AI's mean criterion-level score on its held-out legal-agent evaluation.LOCA-Bench 256K leaderboardTests long-context agents as tool environments grow to a 256K-token environment description.Market-Bench leaderboardA quantitative-trading implementation benchmark that asks models to build backtesters under market-book liquidity and execution-delay constraints, then compares their outputs with a verifier.MazeBench leaderboardTests whether a coding agent can explore a 256-room 3D puzzle world and collect its 100 hidden gems, using box-pushing puzzles, camera rotation, and long-horizon planning.MCP Atlas leaderboardA benchmark for tool-calling over Model Context Protocol integrations and external tools.MCP Mark Verified leaderboardA human-verified edition of MCPMark for MCP tool use across Notion, GitHub, Filesystem, Postgres, and Playwright server environments.MCP-Atlas claim coverage leaderboardAverage coverage of required claims in answers produced during real-world MCP tool-use workflows.MCP-Tasks leaderboardA Model Context Protocol task benchmark used in Qwen's launch tables to measure practical execution over MCP-style tools and integrations.MEWC leaderboardA benchmark that evaluates AI agents on multi-environment web challenges, testing navigation and task completion across diverse live web environments.MLE-Bench Lite leaderboardA lightweight machine-learning competition benchmark that measures whether models can iteratively train, evaluate, and improve ML systems in low-resource settings.MM-ClawBench leaderboardAn OpenClaw-derived agent benchmark covering practical work and life tasks such as office document delivery, research, planning, and code maintenance.MobileWorld leaderboardA mobile-use agent benchmark for completing interactive tasks in smartphone environments.OpenHands Index leaderboardA holistic coding-agent benchmark that evaluates AI agents across issue resolution, frontend work, greenfield development, testing, and information gathering.OSWorld leaderboardA computer-use benchmark for GUI task completion across the broader OSWorld task suite.OSWorld 2.0 leaderboardOSWorld 2.0 evaluates long-horizon computer-use tasks.OSWorld-Verified leaderboardOSWorld-Verified is the July 2025 repaired release of OSWorld's real-computer evaluation.PinchBench leaderboardAn OpenClaw agent benchmark from Kilo that measures successful task completion across standardized real-world agent workflows.QwenClawBench leaderboardQwen's internal OpenClaw-style benchmark for measuring broad real-world agent performance across practical productivity and research tasks.QwenWebBench leaderboardA Qwen benchmark for artifact and webpage generation quality reported as an Elo-style rating.ResearchClawBench leaderboardAn end-to-end autonomous scientific research benchmark with 40 tasks across 10 scientific domains, where agents receive related literature and raw data, then attempt to rediscover the hidden target paper.ReviewBench leaderboardMeasures how reliably code-review agents identify useful issues in real pull requests while avoiding false positives.SpreadsheetBench 2 leaderboardA spreadsheet-focused benchmark for agentic analysis and editing workflows.SWE Refactor Bench leaderboardTests whether coding agents can complete long-horizon, whole-repository stack migrations while preserving the original program's behavior.SWE-Atlas Refactoring leaderboardA Scale SWE-Atlas software-engineering agent benchmark focused on refactoring tasks.SWE-sweep leaderboardTests whether coding agents can find and fix multiple bugs across a repository without issue descriptions, while preserving existing behavior.TAU-bench leaderboardOriginal TAU-bench tests an agent's tool use and policy compliance in simulated airline and retail conversations.Terminal-Bench 2.0 leaderboardA benchmark for agentic software engineering tasks executed in real terminal environments.Terminal-Bench 2.1 leaderboardTerminal-Bench 2.1 results, mostly provider-reported, stored separately from the Terminal-Bench 2.0 lane.Terminal-Bench 2.1 (Vals) leaderboardVals AI’s independent run of the Terminal-Bench 2.1 terminal-task suite with published easy, medium, and hard splits.Terminal-Bench 3.0 leaderboardA continuously maintained benchmark for difficult computer work, including coding, deep learning, finance, engineering, math, and science tasks.Terminal-Bench 4.0 leaderboardThe current Terminal-Bench release measures difficult computer work after recalibrating task resources, fixing unstable tasks, and removing tasks that no longer separate frontier systems.Terminal-Bench-Science 0.1 leaderboardA benchmark of AI agents completing expert-curated research workflows across the life, physical, Earth, mathematical, and engineering sciences.Terminal-Bench-Science 0.1 (6x verifier timeout) leaderboardGoogle-run scientific workflow evaluation using a sixfold verifier timeout to address verification timeouts.Toolathlon leaderboardA tool-use benchmark focused on selecting, sequencing, and completing tasks with external tools.Toolathlon Verified avg. turns leaderboardAverage assistant turns per Toolathlon Verified trajectory.Toolathlon Verified Pass@3 leaderboardFraction of Toolathlon Verified tasks solved in at least one of three trials.Toolathlon Verified Pass³ leaderboardFraction of Toolathlon Verified tasks solved in all three independent trials.Toolathlon-Verified leaderboardA verified tool-use benchmark variant for completing multi-step workflows with external tools.VITA-Bench leaderboardAn interactive real-world agent benchmark grounded in practical consumer-service tasks such as delivery, in-store consumption, and online travel workflows.WebArena leaderboardWebArena tests whether a browser-agent system can complete 812 long-horizon tasks inside self-hosted replicas of functional websites.WebArena-Verified leaderboardWebArena-Verified is an audited release of the WebArena browser-agent benchmark.WebVoyager leaderboardA browser-agent benchmark for completing multi-step workflows on live websites.WideResearch leaderboardA broad research-agent benchmark for open-ended information gathering, synthesis, and answer construction across wide search spaces.ZClawBench leaderboardA Z.AI benchmark for OpenClaw-style agent workflows spanning information search, office work, data analysis, development and operations, automation, and security.τ²-bench Airline leaderboardτ²-bench Airline tests conversational agents on airline customer-service tasks governed by domain policy and database-changing tools.τ²-bench results leaderboardThis route is a sourced ledger for published τ²-bench results.τ³-bench results leaderboardτ³-bench is the current evolution of Sierra's tool-agent-user framework, adding corrected task releases and newer knowledge and voice evaluation modes alongside airline, retail, and telecom.

Coding66

external325

Instruction Following6

AA-IFBench leaderboardAn Artificial Analysis IFBench score.Gray Swan IPI (15 attempts) leaderboardMeasures how often indirect prompt-injection attacks succeed within fifteen attempts.IFBench leaderboardIFBench evaluates precise instruction-following generalization on 58 challenging, verifiable out-of-domain constraints.IFEval leaderboardA benchmark of 541 prompts built from 25 verifiable instruction types.MultiChallenge (provider reports) leaderboardTests instruction retention, memory of user information, versioned editing, and coherence over multi-turn conversations.SOB Value Acc leaderboardA structured-output benchmark from Interfaze measuring whether extracted JSON leaf values exactly match verified ground truth.

Knowledge61

korean8

CLIcK leaderboardEvaluates Korean culture and linguistics.HRM8K leaderboardKorean mathematical reasoning (high-school to Olympiad level).KMMLU leaderboardEvaluates Korean expert-level knowledge across 45 subjects.KMMLU-Hard leaderboardA filtered hard subset of KMMLU containing ~5,000 questions that most models get wrong.KMMLU-Pro leaderboardKorean National Professional Licensure exams evaluating professional-grade knowledge.KMMLU-Redux leaderboardCleaned KMMLU from national technical qualification exams, with errors removed, decontaminated, and deduplicated.KoBALT leaderboardEvaluates advanced Korean linguistic competence.Korean CSAT leaderboardThe Korean SAT exam.

Mathematics34

Multilingual15

AA Global-MMLU-Lite leaderboardAn independently evaluated multilingual knowledge result from Artificial Analysis.GMMLU leaderboardMMLU-style knowledge evaluation across 42 high- and low-resource languages.INCLUDE leaderboardA multilingual benchmark used in provider tables to measure inclusive language coverage and cross-lingual understanding beyond common high-resource languages.Long-context translation leaderboardCohere's long-document translation evaluation: two book chapters translated in a single call, scored per paragraph with xCOMET-XL.MAXIFE leaderboardA multilingual instruction-following and understanding benchmark row published in Qwen's launch comparisons.MGSM leaderboardA multilingual benchmark that translates 250 grade school math problems from GSM8K into 10 typologically diverse languages: Bengali, German, Spanish, French, Japanese, Russian, Swahili, Telugu, Thai, and Chinese.MILU leaderboardCulturally grounded knowledge comprehension across ten Indic languages and English.MKQA-11 leaderboardA display-only multilingual QA retrieval benchmark reported by Liquid AI for LFM2.5 retriever models, using Recall@20 across 11 languages.MMLU-ProX leaderboardA multilingual extension of professional-level academic evaluation across many languages.NanoBEIR Multilingual leaderboardA display-only multilingual retrieval benchmark reported by Liquid AI for LFM2.5 retriever models, using NDCG@10 across 11 languages.NOVA-63 leaderboardA broad multilingual benchmark row from Qwen's launch comparisons intended to measure cross-lingual capability beyond a single language family.PolyMath leaderboardA multilingual mathematical reasoning benchmark that tests whether math performance transfers across languages rather than only in English.SWE Multilingual leaderboardA multilingual extension of SWE-bench covering 300 problems across 9 programming languages, testing code generation and bug fixing beyond Python.VWT2k-lite leaderboardA lighter multilingual benchmark slice published in provider tables for broad cross-lingual transfer and understanding.WMT26 leaderboardMachine-translation quality on the WMT26 general translation task, averaged across all evaluated languages and scored 0-100 by an LLM judge.

Multimodal & Grounded72

Reasoning36

AA-LCR leaderboardAn Artificial Analysis long-context reasoning evaluation.AI-Needle leaderboardA long-context retrieval benchmark that measures whether a model can recover relevant information embedded deep inside very long contexts.ARC-AGI-1 leaderboardARC Prize fluid-intelligence benchmark using novel visual grid transformations.ARC-AGI-2 leaderboardARC-AGI-2 tests whether a system can infer a transformation from example grids and apply it to a new grid.ARC-AGI-3 leaderboardAn interactive successor to ARC-AGI-2 that evaluates whether an AI agent can learn unfamiliar task mechanics through action and feedback.ARC-AGI-3 Provider Adapter leaderboardARC Prize verified interactive reasoning results using a provider-specific stateful adapter, kept separate from Standard harness results.BBH leaderboardA suite of 23 challenging tasks from the BIG-Bench collaborative benchmark where prior language models failed to exceed average human performance, even with chain-of-thought prompting.BullshitBench v2 leaderboardA benchmark that tests whether AI models challenge nonsensical, ill-posed, or logically flawed prompts instead of confidently generating incorrect answers.CLUEWSC leaderboardA Chinese Winograd Schema Challenge benchmark reported in DeepSeek-V4 base-model evaluations.Conceptual Reasoning leaderboardTests whether model judgments rank argumentative critiques in the same order as expert human ratings across philosophy, AI alignment, and other concept-heavy texts.CorpusQA 1M leaderboardA million-token CorpusQA long-context question-answering benchmark reported in DeepSeek-V4 model evaluations.CritPt leaderboardAn Artificial Analysis metric for research-level physics reasoning.DROP leaderboardA reading-comprehension benchmark requiring discrete reasoning over paragraphs, reported in DeepSeek-V4 base-model evaluations.GeneBench-Pro leaderboardA multistage statistical-reasoning benchmark for genomics and biological-data analysis agents.GPQA Diamond leaderboardThe hardest subset of GPQA featuring the most challenging graduate-level science questions.Graphwalks BFS 128K leaderboardLong-context graph traversal benchmark using breadth-first search tasks.GraphWalks BFS 256K–1M leaderboardBreadth-first-search graph traversal on 200 problems with context lengths from 256K to 1M tokens.Graphwalks Parents 128K leaderboardLong-context benchmark for recovering parent relationships inside graph tasks.HellaSwag leaderboardA commonsense natural-language inference benchmark reported in DeepSeek-V4 base-model evaluations.JevBench leaderboardTyped-decision evaluation across 1,624 decisions, combining chance-corrected intelligence, probability calibration, latency, and cost.JevBench 1.4 leaderboardDecision-system evaluation combining chance-corrected intelligence, probability calibration, latency, and cost across 842 decisions.JevBench 1.5 leaderboardTyped-decision evaluation across 1,624 decisions, combining chance-corrected intelligence, probability calibration, latency, and cost.LisanBench leaderboardA word-chain reasoning benchmark that tests planning, recall, constraint following, and vocabulary depth by asking models to extend non-repeating edit-distance-1 chains.LongBench v2 leaderboardA long-context benchmark that measures whether models can actually use extended context windows for reasoning and retrieval.MLCR-AA leaderboardAn open benchmark from Wisedocs, run by Artificial Analysis, measuring how well models reason over long, fragmented medical records with multi-document synthesis.MRCR 1M leaderboardA million-token MRCR long-context retrieval benchmark reported in DeepSeek-V4 model evaluations.MRCR v2 128K-256K leaderboardMRCR v2 slice focused on very long contexts at 128K-256K lengths.MRCR v2 256K-512K leaderboardMRCR v2 slice focused on retrieval across 256K-512K-token contexts.MRCR v2 512K-1M leaderboardMRCR v2 slice focused on retrieval across 512K-1M-token contexts.MRCR v2 64K-128K leaderboardMRCR v2 slice focused on long-context retrieval at 64K-128K lengths.MRCRv2 leaderboardA long-context benchmark for memory, retrieval, and multi-round coherence over large contexts.MuSR leaderboardA dataset for evaluating language models on multistep soft reasoning tasks specified in natural language narratives.Pencil Puzzle Bench leaderboardA multi-step verifiable reasoning benchmark that evaluates whether models can solve pencil puzzles with unique solutions.STEMBench leaderboardScores how far a model advances 120 open research tasks across biology, chemistry, physics, astronomy, Earth science and engineering on a 0 to 4 progress scale, crediting supported partial work and honest statements of what remains unresolved.WildBench leaderboardAn automated evaluation framework using 1,000+ real-world user tasks covering reasoning, planning, coding, and creative writing.WinoGrande leaderboardA commonsense coreference benchmark reported in DeepSeek-V4 base-model evaluations.