CWE-Bench-AA Benchmark Leaderboard
CWE-Bench-AA pass@1
Pontuação
CWE-Bench-AA
CWE-Bench-AA: Pontuação vs. custo por tarefa
Uso de tokens
CWE-Bench-AA: tokens de saída por tarefa
Custo
CWE-Bench-AA: custo por tarefa
Velocidade
CWE-Bench-AA: tempo por tarefa
Pontuação vs. data de lançamento
CWE-Bench-AA: pontuação vs. data de lançamento
Perguntas frequentes
O CWE-Bench-AA é a implementação da Artificial Analysis do CWE-bench, um benchmark de cibersegurança defensiva da Collinear AI. Cada tarefa fornece a um agente de programação um checkout de um repositório real de código aberto e uma única instrução para auditar o código e corrigir o que encontrar. A tarefa descreve a área afetada, mas não a localização exata, e o agente nunca é solicitado a criar exploits. O conjunto reúne 120 tarefas reservadas, cobrindo todas as dez categorias OWASP Top 10 (2025), em C/C++, Go, Java, JavaScript/TypeScript, Python e Rust.
Cada tarefa tem um verificador programático determinístico executado quando o agente termina. Ela só conta como resolvida quando o exploit deixa de funcionar e o comportamento legítimo continua funcionando. O verificador retorna 1 ou 0. A pontuação principal é pass@1: a proporção das 120 tarefas resolvidas em uma única tentativa.
Ambos usam as mesmas tarefas reservadas e o mesmo verificador determinístico, mas as execuções diferem em três aspectos. A Collinear AI executa cada modelo no produto de agente do respectivo fornecedor, como Claude Code ou Codex, ou Terminus-2 para modelos de pesos abertos, faz quatro tentativas por tarefa e reporta pass@1 e pass@4. Também registra uma faixa de crédito parcial avaliada por LLM. O CWE-Bench-AA executa todos os modelos em nosso agente Stirrup de código aberto com os mesmos prompts, realiza cada tarefa uma vez e avalia apenas o verificador determinístico. Por isso, as pontuações desta página não são diretamente comparáveis aos resultados publicados pela Collinear AI.
Não. O conjunto de 120 tarefas é reservado e privado da Collinear AI e da Artificial Analysis e não está disponível para nenhuma organização externa, o que ajuda a reduzir o risco de contaminação dos dados de treinamento. A Collinear AI oferece separadamente um corpus de treinamento com mais de 1.000 tarefas que nunca se sobrepõe ao conjunto de avaliação.
Sim. O CWE-Bench-AA é uma das três avaliações do Artificial Analysis Cyber Index, com o mesmo peso que o DeepsecBench-AA e o CyberGym-E2E-AA. Ele não contribui para o Artificial Analysis Intelligence Index.
Explorar avaliações
Artificial Analysis Intelligence Index v4.3.2A composite benchmark aggregating ten challenging evaluations to provide a holistic measure of AI capabilities across mathematics, science, coding, and reasoning.
Artificial Analysis Openness IndexA composite measure providing an industry standard to communicate model openness for users and developers.
AA-Briefcase v1.1: Agentic Knowledge Work BenchmarkA private evaluation developed by Artificial Analysis for frontier agentic capability in long-horizon knowledge work, testing agents on realistic business workflows that require deliverables such as spreadsheets, presentations, and memos.
GDPval-AA v2.1 LeaderboardGDPval-AA v2.1 is Artificial Analysis' evaluation framework for OpenAI's GDPval dataset. It tests AI models on real-world tasks across 44 occupations and 9 major industries. Models are given shell access and web browsing capabilities in an agentic loop via Stirrup to solve tasks, with Elo ratings derived from blind pairwise comparisons.
APEX-Agents-AA Benchmark LeaderboardArtificial Analysis' implementation of the APEX-Agents benchmark, testing AI agents on long-horizon, cross-application tasks in professional-services environments with realistic application tooling.
AA-AnalystAgent Benchmark LeaderboardArtificial Analysis' data analysis benchmark, testing AI agents on their ability to work with spreadsheets and documents to answer quantitative questions a Business Analyst or Data Analyst would face day-to-day.
AutomationBench-AA: Agentic SaaS Workflow BenchmarkA benchmark measuring agentic task completion across simulated SaaS application environments, scoring the share of each task's objectives completed without guardrail violations.
Harvey LAB-AA Benchmark LeaderboardArtificial Analysis' implementation of Harvey's Legal Agent Benchmark (LAB), testing AI agents on real-world legal work from Harvey's dataset of 120 private tasks spanning 24 legal practice areas. The agent reads case documents in a sandbox and produces legal deliverables (e.g., memos, disclosure schedules, deposition summaries), graded criterion-by-criterion by a single LLM rubric judge.
GDP.pdf Benchmark LeaderboardArtificial Analysis' implementation of Surge AI's GDP.pdf benchmark, testing whether language models can reason over long, real-world professional documents and satisfy detailed task-specific criteria.
EnterpriseOps-Gym-AA Benchmark LeaderboardArtificial Analysis' independent implementation of ServiceNow's EnterpriseOps-Gym, an agentic benchmark testing whether LLM agents can complete stateful, multi-step enterprise workflows across eight business domains via live tool use, graded on the final state of the underlying databases.
Terminal-Bench 4.0 Benchmark LeaderboardA harder 66-task benchmark of complex terminal work across software, machine learning, science, operations, security, hardware, and media, with recalibrated compute and time allowances and improved instructions, environments, and verifiers.
Terminal-Bench-Science 0.1 Benchmark LeaderboardA 70-task benchmark of research workflows authored and reviewed by domain experts across the life, physical, earth, mathematical, and engineering sciences, each completed in a terminal and checked by its own set of tests.
Artificial Analysis Long Context Reasoning Benchmark LeaderboardA challenging benchmark measuring language models' ability to extract, reason about, and synthesize information from long-form documents ranging from 10k to 100k tokens (measured using the cl100k_base tokenizer).
AA-Omniscience: Knowledge and Hallucination BenchmarkA benchmark measuring factual recall and hallucination across various economically relevant domains.
SciCode Benchmark LeaderboardA scientist-curated coding benchmark featuring 288 test set subproblems from 80 laboratory problems across 16 scientific disciplines.
Humanity's Last Exam Benchmark LeaderboardA frontier-level benchmark with 2,500 expert-vetted questions across mathematics, sciences, and humanities, designed to be the final closed-ended academic evaluation.
CritPt Benchmark LeaderboardA benchmark designed to test LLMs on research-level physics reasoning tasks, featuring 71 composite research challenges.
The most challenging 198 questions from GPQA, where PhD experts achieve 65% accuracy but skilled non-experts only reach 34% despite web access.
ITBench-AA Benchmark LeaderboardArtificial Analysis' implementation of IBM's ITBench benchmark, testing AI agents on Kubernetes incident root-cause analysis from offline incident snapshots. The agent inspects alerts, events, traces, and topology and identifies the contributing-factor entities (deployments, pods, namespaces, network policies, etc.) responsible for the failure.
MMMU-Pro Benchmark LeaderboardAn enhanced MMMU benchmark that eliminates shortcuts and guessing strategies to more rigorously test multimodal models across 30 academic disciplines.
IFBench Benchmark LeaderboardA benchmark evaluating precise instruction-following generalization on 58 diverse, verifiable out-of-domain constraints that test models' ability to follow specific output requirements.
Medical Long Context Reasoning (MLCR-AA)An open benchmark from Wisedocs measuring how well models reason over long, fragmented medical records, performing the multi-document synthesis claims professionals rely on when reviewing insurance and healthcare cases.
𝜏³-Banking Benchmark LeaderboardA fintech customer-support benchmark from the 𝜏-Knowledge framework that tests whether agents can navigate a large unstructured knowledge base and execute multi-step tool calls to resolve realistic banking workflows.
Terminal-Bench 2.1 Benchmark LeaderboardA verified refresh of Terminal-Bench 2.0 — 89 curated tasks across software engineering, system administration, data processing, model training, and security, with environment and instruction fixes so scores reflect agent capability rather than environment gaps.
Terminal-Bench Hard Benchmark LeaderboardAn agentic benchmark evaluating AI capabilities in terminal environments through software engineering, system administration, and data processing tasks.
𝜏²-Bench Telecom Benchmark LeaderboardA dual-control conversational AI benchmark simulating technical support scenarios where both agent and user must coordinate actions to resolve telecom service issues.
MMLU-Pro Benchmark LeaderboardAn enhanced version of MMLU with 12,000 graduate-level questions across 14 subject areas, featuring ten answer options and deeper reasoning requirements.
LiveCodeBench Benchmark LeaderboardA contamination-free coding benchmark that continuously harvests fresh competitive programming problems from LeetCode, AtCoder, and CodeForces, evaluating code generation, self-repair, and execution.
MATH-500 Benchmark LeaderboardA 500-problem subset from the MATH dataset, featuring competition-level mathematics across six domains including algebra, geometry, and number theory.
AIME 2025 Benchmark LeaderboardAll 30 problems from the 2025 American Invitational Mathematics Examination, testing olympiad-level mathematical reasoning with integer answers from 000-999.
Global-MMLU-Lite Benchmark LeaderboardA lightweight, multilingual version of MMLU, designed to evaluate knowledge and reasoning skills across a diverse range of languages and cultural contexts.