模型测评：Artificial Analysis 指标

公开 LLM 指标聚合展示，包含智能、代码、数学、速度、价格、延迟和性价比等维度。

智能

Anthropic Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)、Anthropic Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)、OpenAI GPT-6 Astra (max) 在 Artificial Analysis Intelligence Index 中排名靠前。

输出速度

Celeris Celeris-1、Inception Mercury 2、Meta Muse Spark 1.3 (max) 的输出速度更高，适合强调响应速度的场景。

代码

Anthropic Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)、Anthropic Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)、Anthropic Claude Fable 5.1 (Adaptive Reasoning, High Effort, Default Fallback) 在代码能力指数中排名靠前。

数学

OpenAI GPT-5.2 (xhigh)、OpenAI GPT-5 Codex (high)、Google Gemini 3 Flash Preview (Reasoning) 在数学能力指数中排名靠前。

智能

Artificial Analysis Intelligence Index · 越高越好

Anthropic Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)

Top 53.4平均 51.3910 项

代码

Artificial Analysis Coding Index · 越高越好

Anthropic Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)

Top 81.6平均 78.3510 项

数学

Artificial Analysis Math Index · 越高越好

OpenAI GPT-5.2 (xhigh)

Top 99平均 96.5510 项

速度

输出 tokens/s · 越高越好

Celeris Celeris-1

Top 2362.6平均 616.4310 项

各维度 Top 3 差距

每个维度以第 1 名为 100，对比第 2、3 名与头名的接近程度。

智能

#1Anthropic Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)53.4

#2Anthropic Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)53.2

#3OpenAI GPT-6 Astra (max)52.8

代码

#1Anthropic Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)81.6

#2Anthropic Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)80.7

#3Anthropic Claude Fable 5.1 (Adaptive Reasoning, High Effort, Default Fallback)79.1

数学

#1OpenAI GPT-5.2 (xhigh)99

#2OpenAI GPT-5 Codex (high)98.7

#3Google Gemini 3 Flash Preview (Reasoning)97

速度

#1Celeris Celeris-12362.6

#2Inception Mercury 2933.5

#3Meta Muse Spark 1.3 (max)454.8

混合价格

#1Meta Llama 3.1 Instruct 8B0.028

#2Google Gemma 4 E4B (Non-reasoning)0.04

#3Google Gemma 4 E4B (Reasoning)0.04

输入价格

#1Google Gemma 4 E4B (Non-reasoning)0.02

#2Google Gemma 4 E4B (Reasoning)0.02

#3Meta Llama 3.1 Instruct 8B0.02

输出价格

#1Meta Llama 3.1 Instruct 8B0.05

#2Google Gemma 4 E4B (Non-reasoning)0.1

#3Google Gemma 4 E4B (Reasoning)0.1

首响延迟

#1Cohere Command A+0.17

#2IBM Granite 4.2 3B0.21

#3IBM Granite 4.2 30B0.25

性价比

#1Meta Llama 3.1 Instruct 8B246.4

#2Sapiens AI Agnes 2.5 Pro Beta234.7

#3Tencent Hy3-preview (Reasoning)227

供应商覆盖

#1OpenAI92

#2Alibaba88

#3Google65

智能 Top 5

直接对比头部 5 个模型的实际分数，比占比图更适合看排名差距。

#1Anthropic Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)53.4

#2Anthropic Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)53.2

#3OpenAI GPT-6 Astra (max)52.8

#4OpenAI GPT-6 Astra (xhigh)52.5

#5Anthropic Claude Fable 5.1 (Adaptive Reasoning, High Effort, Default Fallback)51.2

智能

Artificial Analysis Intelligence Index · 越高越好

#1Anthropic Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)53.4

#2Anthropic Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)53.2

#3OpenAI GPT-6 Astra (max)52.8

#4OpenAI GPT-6 Astra (xhigh)52.5

#5Anthropic Claude Fable 5.1 (Adaptive Reasoning, High Effort, Default Fallback)51.2

#6OpenAI GPT-6 Astra (high)51

#7Anthropic Claude Opus 5 (Adaptive Reasoning, Max Effort)50.7

#8OpenAI GPT-6 Astra (medium)49.7

#9Anthropic Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)49.7

#10Anthropic Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)49.7

显示前 10 项，完整模型名见图表下方列表。

代码

Artificial Analysis Coding Index · 越高越好

#1Anthropic Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)81.6

#2Anthropic Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)80.7

#3Anthropic Claude Fable 5.1 (Adaptive Reasoning, High Effort, Default Fallback)79.1

#4OpenAI GPT-5.6 Sol (xhigh)78.3

#5Anthropic Claude Opus 5 (Adaptive Reasoning, Max Effort)78

#6OpenAI GPT-5.6 Sol (max)77.4

#7OpenAI GPT-5.6 Sol (high)77.2

#8OpenAI GPT-6 Astra (high)77.1

#9Anthropic Claude Fable 5.1 (Adaptive Reasoning, Medium Effort, Default Fallback)77.1

#10Anthropic Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)77

显示前 10 项，完整模型名见图表下方列表。

数学

Artificial Analysis Math Index · 越高越好

#1OpenAI GPT-5.2 (xhigh)99

#2OpenAI GPT-5 Codex (high)98.7

#3Google Gemini 3 Flash Preview (Reasoning)97

#4OpenAI GPT-5.2 (medium)96.7

#5DeepSeek V3.2 Speciale96.7

#6Xiaomi MiMo-V2-Flash (Reasoning)96.3

#7OpenAI GPT-5.1 Codex (high)95.7

#8Google Gemini 3 Pro Preview (high)95.7

#9Z AI GLM-4.7 (Reasoning)95

#10Kimi K2 Thinking94.7

显示前 10 项，完整模型名见图表下方列表。

速度

输出 tokens/s · 越高越好

#1Celeris Celeris-12362.6

#2Inception Mercury 2933.5

#3Meta Muse Spark 1.3 (max)454.8

#4Multiverse Computing HyperNova 60B 2605 (high, based on gpt-oss-120b)371.8

#5Google Gemini 3.7 Flash (medium)353.4

#6Arcee AI Trinity Large Thinking351.4

#7Google Gemini 3.5 Flash-Lite349.9

#8Google Gemini 3.7 Flash (high)333.3

#9Google Gemini 3.8 Flash (high)330.6

#10Meta Muse Spark 1.3 (xhigh)323

显示前 10 项，完整模型名见图表下方列表。

混合价格

每 100 万 tokens 混合价格 · 越低越好

#1Meta Llama 3.1 Instruct 8B0.028

#2Google Gemma 4 E4B (Non-reasoning)0.04

#3Google Gemma 4 E4B (Reasoning)0.04

#4Sarvam 30B (high)0.047

#5IBM Granite 4.2 3B0.052

#6Alibaba Qwen3.5 4B (Reasoning)0.06

#7Alibaba Qwen3.5 4B (Non-reasoning)0.06

#8Amazon Nova Micro0.061

#9IBM Granite 4.1 8B0.063

#10Multiverse Computing HyperNova 60B 2605 (high, based on gpt-oss-120b)0.065

显示前 10 项，完整模型名见图表下方列表。

输入价格

每 100 万输入 tokens · 越低越好

#1Google Gemma 4 E4B (Non-reasoning)0.02

#2Google Gemma 4 E4B (Reasoning)0.02

#3Meta Llama 3.1 Instruct 8B0.02

#4Sarvam 30B (high)0.026

#5IBM Granite 4.2 3B0.03

#6Alibaba Qwen3.5 4B (Reasoning)0.03

#7Alibaba Qwen3.5 4B (Non-reasoning)0.03

#8IBM Granite 3.3 8B (Non-reasoning)0.03

#9Amazon Nova Micro0.035

#10NVIDIA Nemotron Nano 9B V2 (Reasoning)0.04

显示前 10 项，完整模型名见图表下方列表。

输出价格

每 100 万输出 tokens · 越低越好

#1Meta Llama 3.1 Instruct 8B0.05

#2Google Gemma 4 E4B (Non-reasoning)0.1

#3Google Gemma 4 E4B (Reasoning)0.1

#4Mistral Ministral 3 3B0.1

#5IBM Granite 4.1 8B0.1

#6Sarvam 30B (high)0.11

#7IBM Granite 4.2 3B0.12

#8Amazon Nova Micro0.14

#9Multiverse Computing HyperNova 60B 2605 (high, based on gpt-oss-120b)0.14

#10Meta Llama 3 Instruct 8B0.145

显示前 10 项，完整模型名见图表下方列表。

首响延迟

Median time to first token · 越低越好

#1Cohere Command A+0.17

#2IBM Granite 4.2 3B0.21

#3IBM Granite 4.2 30B0.25

#4Cohere North Mini Code0.25

#5IBM Granite 4.2 8B0.26

#6Google Gemma 4 E4B (Non-reasoning)0.3

#7Cohere Command A0.32

#8Cohere Tiny Aya Global0.32

#9Microsoft Phi-4 Mini Instruct0.37

#10Multiverse Computing HyperNova 60B 2605 (high, based on gpt-oss-120b)0.38

显示前 10 项，完整模型名见图表下方列表。

性价比

智能指数 / 混合价格 · 越高越好

#1Meta Llama 3.1 Instruct 8B246.4

#2Sapiens AI Agnes 2.5 Pro Beta234.7

#3Tencent Hy3-preview (Reasoning)227

#4InclusionAI Ling 3.0 Flash224.3

#5Google Gemma 4 E4B (Reasoning)222.5

#6Alibaba Qwen3.5 4B (Reasoning)218.3

#7Google Gemma 4 E4B (Non-reasoning)187.5

#8Multiverse Computing HyperNova 60B 2605 (high, based on gpt-oss-120b)180

#9Alibaba Qwen3.5 4B (Non-reasoning)180

#10Z AI GLM-5.3-Flash176.8

显示前 10 项，完整模型名见图表下方列表。

供应商覆盖

当前 API 返回的模型数量 · 越高代表覆盖更多

#1OpenAI92

#2Alibaba88

#3Google65

#4Anthropic48

#5DeepSeek35

#6Mistral32

#7SpaceXAI26

#8Z AI22

#9Meta22

#10NVIDIA19

显示前 10 项，完整模型名见图表下方列表。