Public leaderboards
Every public benchmark, in one place
52 benchmarks and leaderboards from labs and universities, explained in plain words — with their latest results copied here (47 so far) and the source stated on every page.
#General knowledge & reasoning 6
Broad tests of expert knowledge, scientific reasoning and novel problem solving.
-
ARC-AGI (ARC Prize)
Puzzles that most people find easy but that require spotting a brand-new pattern rather than recalling facts; the newest version (ARC-AGI-3) drops the AI into small interactive games it has never seen. It is a popular test of whether AI can genuinely learn new skills on the fly.
Leader: GPT-6 Astra - Provider Adapter (High) 99.9
-
CritPt
Unpublished research-level physics problems written by working physicists, in areas like quantum physics and astrophysics. It tests whether AI can do the kind of reasoning needed in real scientific research.
Leader: Gemini 4 Carbon (High) 36.0
-
EQ-Bench
Tests how well AI models handle emotionally sensitive conversations with simulated people who have different personalities and needs.
Leader: claude-opus-5 1364.9
-
Humanity's Last Exam (HLE)
About 2,500 extremely hard questions written by experts in over 100 subjects, from classics to quantum physics, meant to sit at the edge of human knowledge. It is one of the hardest general-knowledge exams for AI.
Leader: GPT 6 Astra 54.8
-
LiveBench
A set of regularly refreshed tests (agent tasks, reasoning, coding, research, design, etc.) designed so AI models cannot have seen the questions beforehand.
Leader: claude-fable-5.2 84.9
-
SimpleBench
Trick questions about everyday situations, space, time and people that ordinary adults usually get right but that can trip up AI. It tests common sense rather than book knowledge.
Leader: Claude Fable 5.2 (unknown) 94.9
#Math 3
Competition, research-level and formal mathematics, from final-answer problems to proofs.
-
FrontierMath
Hundreds of brand-new, unpublished math problems written by professional mathematicians, ranging from hard competition level up to problems that take experts days. Because the problems are secret, it is a strong check of real mathematical ability rather than memorization.
Leader: Claude Fable 5.2 (max) 95.1
-
MathArena
Tests AI models on fresh, hard math problems taken from new competitions and research papers, so the answers cannot have been memorised.
Leader: Gemini 4 Carbon (high) 94.28% ± 3.64%
-
PutnamBench
Tests whether AI systems can write machine-checkable formal proofs for problems from the Putnam, a famously hard undergraduate math competition.
Leader: 1 672
#Coding & software engineering 4
Code generation, bug fixing in real repositories and terminal-based engineering work.
-
SWE Atlas
Tests whether AI coding assistants can do everyday professional software work beyond fixing bugs: answering questions about a codebase, writing tests and refactoring code.
Leader: Fable 5.2 80.55
-
SWE-Bench Pro
Tests whether AI agents can solve long, realistic software tasks in real open-source projects, harder than older bug-fixing tests.
Leader: Opus 5 (Claude Code) xhigh 98.00
-
SWE-rebench
Tests AI coding tools on fresh real-world bug-fix tasks collected after models were trained, so they cannot have seen the answers.
Leader: Anthropic Fable 5 [high] 64.5
-
Terminal-Bench
Tests whether AI agents can complete real tasks by typing commands in a computer terminal, like a developer or system administrator would.
Leader: Opus 5.5 ( max ) 65.8% ± 3.1%
#Agents, tool use & computer use 10
Multi-step tasks where a model calls tools, browses the web or operates a computer.
-
APEX-Agents
Tests AI agents on long professional tasks like those of bankers, consultants and lawyers.
Leader: Gemini 4 Carbon (High) 84.2% ±4.3%
-
GDPval-AA
Tests AI on real office-style professional tasks across many occupations, judged head-to-head.
Leader: Claude Opus 5.5 (Max, Default Fallback) 1866
-
MCPMark
Tests AI agents on tasks that require using MCP tool servers such as Notion and GitHub.
Leader: kimi-k3-max 96.06
-
METR Time Horizons
Measures how long a task (in human working time) an AI can complete about half the time; shows how fast agents are improving.
Leader: claude_mythos_preview_early_inspect 1044.8 min (508.9-3304.3)
-
OSWorld (OSWorld-Verified)
Gives an AI control of a real computer (mouse, keyboard, screen) and asks it to do everyday jobs in apps like a browser, spreadsheet or image editor. It shows how close AI is to operating a PC the way a person does.
Leader: Intelligence-Indeed Agent 90.2
-
OSWorld 2.0
A much harder follow-up to OSWorld: the AI must carry out long, multi-step office and professional workflows on a real computer, jobs that take a person about an hour and a half. It shows whether AI can handle long, realistic computer work rather than quick clicks.
Leader: Claude Opus 5 44.33
-
Remote Labor Index (RLI)
Tests whether AI can complete real paid freelance projects well enough that a client would accept the work.
Leader: Fable 5.2 (NEW) 41.25
-
ScreenSpot-Pro
Shows an AI a screenshot of professional software (like CAD or video editors) and asks it to point to the right button or text. It tests whether an AI can "see" where to click, a basic step for computer-using agents.
Leader: GroundCascade (Indeed-UI-32B + KV-Ground-8B) 85.14
-
Toolathlon (Tool Decathlon)
Tests AI agents on long, multi-step tasks across many everyday software tools.
Leader: GLM 5.3 Flash (max) 78.4 ± 1.9
-
Vending-Bench 2
An AI runs a simulated vending-machine business for a long time; tests staying sensible over many days.
Leader: Gemini 4 Carbon $16,484.21 ± $1,096
#Long context 1
How well models find, track and reason over information in very long inputs.
#Instruction following 1
Whether models obey explicit constraints and keep track of instructions across turns.
#Hallucination & factuality 2
How often models state false facts, invent details or fail to stay grounded in sources.
-
AA-Omniscience
Tests how much factual knowledge an AI really has across 42 topics, and punishes it for confidently making things up instead of saying 'I don't know'. A score of 0 means as many wrong as right answers.
Leader: Claude Fable 5.2 (Max, Default Fallback) 67
-
Vectara Hallucination Leaderboard
Asks AI models to summarize short documents and uses an automated checker to count how often the summary adds things that are not in the original text. It shows how likely a model is to make things up when it should just be faithful to a source.
Leader: antgroup/finix_s1_32b 1.8 %
#Human-preference arenas 3
Leaderboards built from blind side-by-side votes by real users.
-
Arena (formerly LMArena / Chatbot Arena)
Real people type a question, get two anonymous answers from different AI chatbots, and vote for the better one. Millions of these votes are turned into a ranking of which chatbots people prefer.
Leader: Gemini 4 Carbon (High) 1526
-
Design Arena
People ask several AI models to build the same thing, such as a website, game or slide deck, then vote on which result looks and works better. The votes become a ranking of which models are best at making designs and apps.
Leader: Claude Opus 5.5 (high) 1445
- Artificial Analysis Image & Video Arenas People are shown two AI-made pictures (or videos) for the same request without knowing which model made them, and pick the better one. The votes become a ranking of image and video generators.
#Multimodal & vision 1
Understanding and reasoning over images, charts, diagrams and video.
#OCR & document understanding 1
Reading text, tables, formulas and layout from scanned and digital documents.
#Speech recognition (ASR) 2
Speech-to-text accuracy and speed, usually measured by word error rate.
-
Open ASR Leaderboard
Tests how accurately speech-to-text systems write down what people say in English recordings (meetings, earnings calls, audiobooks and more). Lower error means fewer wrong words.
Leader: zoom/scribe_v2.1_pro 4.87
- Artificial Analysis Speech to Text Compares services that turn spoken audio into written text: how many words they get wrong, how fast they are and what they cost.
#Text-to-speech 2
Quality and naturalness of synthetic speech, mostly judged by blind listening votes.
-
TTS Arena V2
People listen to two computer-generated voices reading the same text and pick the one that sounds better. The ranking shows which text-to-speech systems people actually prefer.
Leader: Aurora 1687
- Artificial Analysis Speech Arena Listeners hear the same sentence read by two AI voices without knowing which is which and pick the more natural one. The votes become a ranking of text-to-speech models.
#Safety & security 3
Resistance to jailbreaks and prompt injection, hazardous knowledge and cyber capabilities.
-
CyberGym
Gives AI agents a description of a real security bug in open-source software plus the code, and asks them to write an input that actually triggers the bug. It shows how good AI has become at the hands-on work of finding and reproducing vulnerabilities, which helps defenders and could help attackers.
Leader: Creation Model 90.6%
-
F5 Labs CASI & ARS Leaderboard
Tests how hard it is to trick AI models into harmful behavior using attack prompts. A higher score means the model is harder to break.
Leader: Claude Sonnet 5 98.00
-
Scale Fortress
Tests whether AI models can be pushed into giving dangerous help (e.g. weapons or crime) by adversarial prompts. Lower is safer.
Leader: claude-opus-5-5-max 1
#Multilingual 1
Knowledge and understanding across many languages, not just English.
#Polish language 3
Benchmarks and leaderboards built specifically for Polish.
-
CPTU-Bench
Checks how well AI models really understand Polish: reading between the lines (sarcasm, hidden meaning, idioms) and answering tricky questions with logic traps, ambiguity or absurd premises. It shows whether a model grasps Polish beyond the literal words.
Leader: meta-models/Muse-Glimmer-30B reasoning (API, FP8) 4.42
-
PL-MTEB (Polish MTEB)
Tests how well "embedding" models (the part of AI search that turns text into numbers so similar meanings can be found) work on Polish text, on tasks like finding similar sentences, sorting reviews and grouping documents.
Leader: codefuse-ai/F2LLM-v2-14B 75.13
-
Polish Linguistic and Cultural Competency (PLCC)
600 hand-written questions about Polish history, geography, culture and traditions, art and entertainment, grammar and vocabulary. It shows how well an AI model knows Poland and the Polish language, not just how fluent it sounds.
Leader: Gemini-4-Carbon 98.33
#Speed, price & providers 2
Throughput, latency, cost and usage of models, API providers and inference hardware.
-
OpenRouter Rankings
Shows which AI models people actually use most on OpenRouter, a service that lets developers switch between many AI models. It reflects real-world popularity, not test scores.
Leader: deepseek/deepseek-v4.1-flash-20261017 61170.1
- Artificial Analysis API Provider Benchmarks Compares the companies that host AI models on speed, price and delay, so you can pick the fastest or cheapest provider for the same model.
#Aggregated leaderboards 7
Sites that combine many benchmarks into one place or one index.
-
Epoch AI Benchmarking Hub
Epoch AI runs AI models itself on a set of hard tests (maths, science, coding) and publishes the results in one place, so you can compare models on tests the labs did not grade themselves.
Leader: claude-fable-5-2_max 95.1
-
Epoch Capabilities Index (ECI)
One combined score for how capable an AI model is overall, built by stitching together results from many different tests so models from different years can be compared on one scale.
Leader: Claude Fable 5.2 173.44
-
Kagi LLM Benchmarking Project
Kagi, a search company, regularly quizzes AI models on a private set of tricky reasoning, coding and language questions that change over time, to see which ones actually think well.
Leader: claude-5-fable-thinking 91.4
-
LLM Stats
A website that gathers public test results, prices and speeds for hundreds of AI models and ranks them with a single combined score.
Leader: Claude Fable 5.2 60.8
-
Scale SEAL Leaderboards
Scale AI tests AI models on private, expert-written questions that models cannot have memorised, across areas like coding, reasoning and safety. One representative board (Humanity's Last Exam, Diamond) is shown.
Leader: claude-fable-5-2 72.4
-
Vals AI
Vals AI tests AI models on realistic professional work in finance, law, tax and coding and combines the results into one index, aiming to show how useful models are for real jobs.
Leader: Claude Fable 5.2 71.31
- Artificial Analysis Intelligence Index Artificial Analysis combines results from a bundle of tough tests into one 'intelligence' score per AI model, alongside price and speed, to help people choose a model.