智能Artificial Analysis Intelligence IndexArtificial Analysis Intelligence Index v4.3.2 incorporates 10 evaluations: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, AA-LCR v1.127 of 709 modelsAdd model from specific providerArtificial Analysis Intelligence IndexArtificial Analysis Intelligence Index v4.3.2 includes: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, AA-LCR v1.1. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.
开放权重 / 专有推理 / 非推理仅文本 / 多模态输入Artificial Analysis Intelligence Index：开放权重与专有模型Artificial Analysis Intelligence Index v4.3.2 incorporates 10 evaluations: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, AA-LCR v1.127 of 709 modelsAdd model from specific provider专有开放权重开放权重（商业用途受限）Artificial Analysis Intelligence IndexArtificial Analysis Intelligence Index v4.3.2 includes: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, AA-LCR v1.1. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.
Open WeightsIndicates whether the model weights are available. Models are labelled as 'Commercial Use Restricted' if commercial use is limited by conditions, and as 'Non-commercial' if the license prohibits commercial use.
Intelligence Index 与每任务成本Intelligence Index 与每任务时间Intelligence Index 与每任务输出 Token 数Intelligence Index 与每项任务的成本Artificial Analysis Intelligence Index · Weighted average cost (USD) per Artificial Analysis Intelligence Index task27 of 709 modelsMost attractive quadrantPareto lineDeepSeekOpenAIXiaomiAnthropicSpaceXAIKimiZ AIGoogleMetaStepFunInclusionAIAlibabaAA-Briefcase v1.1GDPval-AA v2.1单任务成本（美元，对数刻度）
查看所有评估Cost per Intelligence Index TaskWeighted average cost per Intelligence Index task. Each evaluation’s cost is calculated from input, cache hit, cache write, reasoning, and answer token prices, divided by task count, and weighted by its Intelligence Index weight.
Artificial Analysis Intelligence IndexArtificial Analysis Intelligence Index v4.3.2 includes: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, AA-LCR v1.1. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.
能力指数衡量模型在特定能力和行业中的表现
财务与会计战略与运营法律医疗与健康工程经济学Artificial Analysis Finance & Accounting IndexIncorporates 7 evaluations: AA-Omniscience, GDPval-AA v2.1, AA-Briefcase v1.1, Humanity's Last Exam, AutomationBench-AA, AA-LCR v1.1, GDP.pdf · Higher is better27 of 227 modelsAdd model from specific provider基准测试EvaluationsEvaluations measured independently by Artificial Analysis · AA-PersonalityBench scores are percentile ranks from 1 to 100编程智能体工具使用私有数据集用户交互金融医疗法律Intelligence Index性格长上下文多模态指令遵循忠实性写作商业查看更多24 of 37 evaluations27 of 709 modelsAdd model from specific providerAA-Briefcase v1.1Agentic knowledge work, Elo
AA-Omniscience AccuracyKnowledge
AA-Omniscience Non-Hallucination Rate1 - hallucination rate
AutomationBench-AAAgentic SaaS workflows
GDPval-AA v2.1Agentic real-world work tasks, Elo
Humanity's Last ExamReasoning & knowledge
GDP.pdfProfessional document reasoning, All-pass
Terminal-Bench 4.0Agentic coding & terminal use
AA-LCR v1.1Long context reasoning
SciCodeUnder reviewCoding
CritPtPhysics reasoning
Terminal-Bench-Science 0.1NewAgentic scientific research workflows in a terminal
FrontierCode 1.1NewAgentic software engineering in open-source repositories
AA-AnalystAgentQuantitative analysis on spreadsheets & documents
Harvey LAB-AA v1.1UpdatedLegal agentic work, Hallucination-Gated All-Pass Rate
EnterpriseOps-Gym-AAAgentic business operations
AA-3DNewAgentic interactive 3D experiences, Elo
ITBench-AAKubernetes incident root-cause analysis
MLCR-AAUpdatedMedical long context reasoning
MMMU-ProVisual reasoning
AA-PersonalityBench: Expressiveness ScoreNewRich, varied, well-crafted language
AA-PersonalityBench: Organization ScoreNewWell-structured, scannable, answer-first
AA-PersonalityBench: Transparency ScoreNewTruthful, acknowledges uncertainty, non-sycophantic
AA-PersonalityBench: Slop ScoreNewUse of filler and generic language
Intelligence Evaluation RelevanceWhile model intelligence generally translates across use cases, specific evaluations may be more relevant for certain use cases.
Artificial Analysis Intelligence IndexArtificial Analysis Intelligence Index v4.3.2 includes: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, AA-LCR v1.1. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.
AA-Briefcase v1.1AA-Briefcase EloAA-Briefcase 评分细则得分（%）分析质量与呈现 EloAA-Briefcase EloAA-Briefcase v1.1 is an agentic knowledge work benchmark developed by Artificial Analysis. AA-Briefcase Elo is a combined metric that aggregates rubric pass rate, analytical quality Elo and presentation Elo · Higher is better27 of 237 modelsAdd model from specific providerAA-Briefcase EloAA-Briefcase Elo is a combined metric that aggregates analytical quality Elo, presentation Elo, and rubric pass rate, with rubric performance converted into Elo via synthetic head-to-head matches. Elo and 95% confidence interval bounds are clamped at 0.
AA-OmniscienceAA-Omniscience IndexAA-Omniscience 准确率AA-Omniscience 幻觉率AA-Omniscience IndexAA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct.27 of 582 modelsAdd model from specific providerAA-Omniscience IndexAA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct.
AA-PersonalityBenchNew个性指数个性特征沟通特征性格指数每个轴对应 8 个性格指数之一：其特质 Elo z 分数的平均值，在基准测试中所有已评估模型之间换算为 1 到 100 的百分位排名16 of 40 modelsDeepSeek V4.1 Flash (max)MiMo-V2.6-ProGPT-6 Astra (max)Claude Opus 5.5 (max with fallback)Claude Fable 5.1 (max with fallback)Grok 4.7 (xhigh)GPT-6.1 Sol (max)GLM-5.3 (max)Gemini 4 Argon (high)DeepSeek V4.1 Flash 1017 (max)Muse Wonder (max)Step 5 PreviewKimi K3 (max)GLM-5.3-FlashMuse Spark Open (xhigh)K2 Horizon 375B A23BOpenness IndexOpenness IndexOpenness Index 构成开放性与智能Artificial Analysis Openness Index：得分Openness Index 以 0 至 100 的标准化评分衡量模型的开放程度（分数越高，越开放）17 of 335 modelsAdd model from specific providerIntelligence Index 比较Intelligence Index 与每任务成本Intelligence Index 与每任务时间Intelligence Index 与输出速度Intelligence Index 与端到端响应时间Intelligence Index 与每项任务的成本Artificial Analysis Intelligence Index · Weighted average cost (USD) per Artificial Analysis Intelligence Index task27 of 709 modelsMost attractive quadrantPareto lineDeepSeekOpenAIXiaomiAnthropicSpaceXAIKimiZ AIGoogleMetaStepFunInclusionAIAlibabaAA-Briefcase v1.1GDPval-AA v2.1单任务成本（美元，对数刻度）
查看所有评估Cost per Intelligence Index TaskWeighted average cost per Intelligence Index task. Each evaluation’s cost is calculated from input, cache hit, cache write, reasoning, and answer token prices, divided by task count, and weighted by its Intelligence Index weight.
Artificial Analysis Intelligence IndexArtificial Analysis Intelligence Index v4.3.2 includes: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, AA-LCR v1.1. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.
Token 使用量每任务输出 Token 数Intelligence Index 与每任务输出 Token 数Intelligence Index Token 使用量Intelligence Index 与 Token 使用量Intelligence Index 每项任务的输出 Token 数运行 Artificial Analysis Intelligence Index 单项任务所用输出 token 数的加权平均值27 of 709 models答案推理Output Tokens per Intelligence Index TaskThe number of tokens required per Intelligence Index task. This is calculated by multiplying the output tokens per eval by the relative weights of each benchmark in the Intelligence Index, then dividing by task count (excluding repeats).
成本每任务成本Intelligence Index 与每任务成本评测明细每项 Intelligence Index 任务的成本Weighted average cost (USD) per Intelligence Index task · Segmented by token type · Lower is better27 of 709 models答案推理缓存写入缓存命中输入Cost per Intelligence Index TaskWeighted average cost per Intelligence Index task. Each evaluation’s cost is calculated from input, cache hit, cache write, reasoning, and answer token prices, divided by task count, and weighted by its Intelligence Index weight.
Intelligence Index 总成本Intelligence Index 与总成本运行 Artificial Analysis Intelligence Index 的成本Cost (USD) to run all evaluations in the Artificial Analysis Intelligence Index27 of 709 modelsAdd model from specific provider输出推理缓存写入缓存读取非缓存输入Cost to Run Artificial Analysis Intelligence IndexThe cost to run the evaluations in the Artificial Analysis Intelligence Index, calculated using the model's input, cache hit, cache write, reasoning, and answer token prices and the number of tokens used across evaluations (excluding repeats).
缓存命中、输入和输出价格混合价格混合价格（堆叠）缓存折扣Intelligence Index 与价格Intelligence Index 与价格（对数、倒置）图像输入价格价格：缓存命中、输入和输出Price (USD per M Tokens)27 of 709 modelsAdd model from specific provider缓存命中输入输出Cache HitPrice per token for cached prompts (previously processed), typically offering a significant discount compared to regular input price, represented as USD per million tokens. The values shown here are the cache hit price; cache write and cache storage are billed separately and vary by provider — see "Cache pricing by provider" for detail.
4 more notes上下文窗口上下文窗口Intelligence Index 与上下文窗口上下文窗口Context window: tokens limit · Higher is better27 of 709 modelsAdd model from specific providerContext Window for RAGLarger context windows are relevant to RAG (Retrieval Augmented Generation) LLM workflows which typically involve reasoning and information retrieval of large amounts of data.
Context WindowMaximum number of combined input & output tokens. Output tokens commonly have a significantly lower limit (varied by model).
速度按输出速度（每秒 token 数）衡量
输出速度按提示词类型划分的输出速度输出速度波动输出速度随时间的变化输出速度与价格延迟与输出速度输出速度Output tokens per second · Higher is better27 of 709 modelsAdd model from specific providerOutput SpeedTokens per second received while the model is generating tokens (ie. after first chunk has been received from the API for models which support streaming).
Model Performance RepresentationFigures represent performance of the model's first-party API or the median across providers where a first-party API is not available.
每任务时间Intelligence Index 与每任务时间成本与每任务时间Intelligence Index 每项任务耗时Weighted average decode time (minutes) per Intelligence Index task · Excludes TTFT and overhead time · Lower is better27 of 709 modelsTime per Intelligence Index TaskThe weighted average time (seconds) per Artificial Analysis Intelligence Index task. This is calculated by dividing output tokens per task by output speed, weighted by the relative weights of each benchmark in the Intelligence Index.
延迟按首 Token 延迟（秒）衡量
首个答案 Token 延迟首 Token 延迟按提示词类型划分的延迟延迟波动延迟随时间的变化延迟: 首个答案 Token 延迟收到首个回答 token 的秒数 · 包含推理模型的“思考”时间27 of 709 modelsAdd model from specific provider思考（推理模型，如适用）输入处理Time to First Answer TokenTime to first answer token received, in seconds, after API request sent. For reasoning models, this includes the 'thinking' time of the model before providing an answer. For models which do not support streaming, this represents time to receive the completion.
端到端响应时间Seconds to output 500 tokens, calculated based on time to first token, 'thinking' time for reasoning models, and output speed
端到端响应时间按提示词类型划分的端到端响应时间端到端响应时间随时间的变化端到端响应时间Seconds to output 500 tokens, including reasoning model 'thinking' time · Lower is better27 of 709 modelsAdd model from specific provider输出时间“思考”时间（推理模型）输入处理时间End-to-End Response TimeSeconds to receive a 500 token response. Key components:
Input time: Time to receive the first response tokenThinking time (only for reasoning models): Time reasoning models spend outputting tokens to reason prior to providing an answer. Amount of tokens based on the average reasoning tokens across a diverse set of 60 prompts (methodology details).Answer time: Time to generate 500 output tokens, based on output speedModel Performance RepresentationFigures represent performance of the model's first-party API or the median across providers where a first-party API is not available.
模型规模（仅开放权重模型）总参数与活跃参数Intelligence Index 与活跃参数Intelligence Index 与总参数模型规模：总参数量与活跃参数量Comparison between total model parameters and parameters active during inference (billions)27 of 709 modelsAdd model from specific provider活跃参数非活跃参数Total ParametersThe total number of trainable weights and biases in the model, expressed in billions. These parameters are learned during training and determine the model's ability to process and generate responses.
Active Parameters at Inference TimeThe number of parameters actually executed during each inference forward pass, expressed in billions. For Mixture of Experts (MoE) models, a routing mechanism selects a subset of experts per token, resulting in fewer active than total parameters. Dense models use all parameters, so active equals total.