Amazon Bedrock:模型智能、性能与价格

Amazon Bedrock
Amazon Bedrock

分析 Amazon Bedrock 各模型的关键指标,包括质量、价格、输出速度、延迟、上下文窗口等。 本分析旨在帮助你根据使用场景,选择 Amazon Bedrock 提供的最佳模型。

最智能

#1
Claude Opus 5 (max)
Claude Opus 5 (max)
61
#2
Claude Opus 5 (xhigh)
Claude Opus 5 (xhigh)
60
#3
GPT-5.6 Sol (max)
GPT-5.6 Sol (max)
59
#4
Claude Opus 5 (high)
Claude Opus 5 (high)
59
#5
Claude Opus 5 (medium)
Claude Opus 5 (medium)
56

Intelligence Index

共 95 个模型

速度最快

#1
Ministral 3 3B
Ministral 3 3B
408 t/s
#2
Nova Micro
Nova Micro
282 t/s
#3
Ministral 3 8B
Ministral 3 8B
251 t/s
#4
Llama 4 Maverick
Llama 4 Maverick
220 t/s
#5
NVIDIA Nemotron Nano 12B v2 VL
NVIDIA Nemotron Nano 12B v2 VL
210 t/s

输出速度

共 95 个模型

价格最低

#1
Nova Micro
Nova Micro
$0.03
#2
Gemma 3 4B
Gemma 3 4B
$0.04
#3
Nova Lite
Nova Lite
$0.05
#4
NVIDIA Nemotron Nano 9B V2
NVIDIA Nemotron Nano 9B V2
$0.08
#5
gpt-oss-20b (high)
gpt-oss-20b (high)
$0.09

每 100 万 token 的混合价格

共 95 个模型

表示推理模型

Amazon 提供 95 个模型,每个模型的智能、性能和价格特征各不相同。 下方对比了各模型的关键指标。

  • 智能方面,Amazon 上表现最好的模型是 Claude Opus 5 (max)(61)、Claude Opus 5 (xhigh)(60)和GPT-5.6 Sol (max)(59)。
  • 输出速度方面,最快的模型是 Ministral 3 3B(408 t/s)、Nova Micro(282 t/s)和Ministral 3 8B(251 t/s)。 各模型之间的速度差异显著,最快与最慢相差 95%。
  • 延迟方面,GPT-5.4 (Non-reasoning)(0.62 秒)、GPT-5.6 Terra (Non-reasoning)(0.70 秒)和Grok 4.3 (Non-reasoning)(0.71 秒) 的首个答案 Token 延迟最低。
  • 价格方面,Nova Micro($0.03)、Gemma 3 4B($0.04)和Nova Lite($0.05) 每 100 万 token 的混合价格最低。 各模型价格最多相差 3.4 倍。
  • 上下文窗口方面,Claude Opus 5 (max)(1M)、Claude Opus 5 (xhigh)(1M)和Claude Opus 5 (high)(1M) 支持 Amazon 上最大的上下文窗口。

亮点

Artificial Analysis Intelligence Index · Higher is better
Output tokens per second · Higher is better
USD per 1M tokens (blended) · Lower is better

智能评测

Artificial Analysis Intelligence Index

Artificial Analysis Intelligence Index v4.1 incorporates 9 evaluations: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR
Reasoning models are indicated by a lightbulb icon

Artificial Analysis Intelligence Index v4.1 includes: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.

Intelligence Evaluations

Intelligence evaluations measured independently by Artificial Analysis · Higher is better

Agentic real-world work tasks, (Elo-500)/2000

Agentic tool use

Agentic coding & terminal use

Coding

Reasoning & knowledge

Scientific reasoning

Physics reasoning

Long context reasoning

Agentic knowledge work, Elo

Agentic SaaS workflows

Legal agentic work, criterion pass rate

Agentic business operations

Instruction following

Long-horizon agentic tasks

Kubernetes incident root-cause analysis

Visual reasoning

Reasoning models are indicated by a lightbulb icon

While model intelligence generally translates across use cases, specific evaluations may be more relevant for certain use cases.

Artificial Analysis Intelligence Index v4.1 includes: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.

Intelligence Index vs. Price

Blended at 7:2:1 (cache-input-output) · USD per 1M tokens (blended)
Most attractive quadrant
Pareto line
Reasoning models are indicated by a lightbulb icon

While higher intelligence models are typically more expensive, they do not all follow the same price-quality curve.

上下文窗口

Context Window

Context window: tokens limit · Higher is better
Reasoning models are indicated by a lightbulb icon

Larger context windows are relevant to RAG (Retrieval Augmented Generation) LLM workflows which typically involve reasoning and information retrieval of large amounts of data.

Maximum number of combined input & output tokens. Output tokens commonly have a significantly lower limit (varied by model).

价格

Intelligence Index vs. Price

Blended at 7:2:1 (cache-input-output) · USD per 1M tokens (blended)
Most attractive quadrant
Pareto line
Reasoning models are indicated by a lightbulb icon

While higher intelligence models are typically more expensive, they do not all follow the same price-quality curve.

性能摘要

Output Speed vs. Price

Output speed: output tokens per second · USD per 1M tokens (blended)
Most attractive quadrant
Pareto line
Reasoning models are indicated by a lightbulb icon

Tokens per second received while the model is generating tokens (ie. after first chunk has been received from the API for models which support streaming).

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

速度

按输出速度(每秒 token 数)衡量

Output Speed

Output tokens per second · Higher is better
Reasoning models are indicated by a lightbulb icon

Tokens per second received while the model is generating tokens (ie. after first chunk has been received from the API for models which support streaming).

Figures represent performance of the model's first-party API (e.g. OpenAI for o1) or the median across providers where a first-party API is not available (e.g. Meta's Llama models).

延迟

按首 Token 延迟(秒)衡量

Latency: Time To First Answer Token

Seconds to first answer token received · Accounts for reasoning model 'thinking' time
Reasoning models are indicated by a lightbulb icon

Time 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

End-to-End Response Time vs. Price

End-to-end response time: end-to-end seconds to output 500 tokens · USD per 1M tokens (blended)
Most attractive quadrant
Pareto line
Reasoning models are indicated by a lightbulb icon

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

进一步分析
Anthropic 标志
Claude Opus 5 (max)
1M
专有
61
$2.34
54
75.74
84.93
--
Anthropic 标志
Claude Opus 5 (xhigh)
1M
专有
60
$1.80
53
39.14
48.57
--
OpenAI 标志
GPT-5.6 Sol (max)
262k
专有
59
$1.67
116
124.18
128.49
--
Anthropic 标志
Claude Opus 5 (high)
1M
专有
59
$1.23
53
23.63
33.05
--
Anthropic 标志
Claude Opus 5 (medium)
1M
专有
56
$0.72
53
13.29
22.74
--
Anthropic 标志
Claude Opus 4.8 (max)
1M
专有
56
$2.01
62
30.88
39.00
--
OpenAI 标志
GPT-5.6 Terra (max)
262k
专有
55
$0.98
170
137.81
140.75
--
OpenAI 标志
GPT-5.5 (xhigh)
272k
专有
55
$2.25
123
48.62
52.67
--
Anthropic 标志
Claude Opus 4.7 (max)
1M
专有
54
$3.72
63
1.79
9.74
--
Anthropic 标志
Claude Sonnet 5 (max)
1M
专有
53
$1.96
70
164.15
171.29
--
OpenAI 标志
GPT-5.5 (high)
272k
专有
53
$1.53
127
20.24
24.18
--
OpenAI 标志
GPT-5.4 (xhigh)
272k
专有
51
$1.76
155
115.29
118.50
--
OpenAI 标志
GPT-5.6 Luna (max)
262k
专有
51
$0.34
202
128.48
130.96
--
Anthropic 标志
Claude Opus 5 (low)
1M
专有
51
$0.43
53
4.70
14.11
--
OpenAI 标志
GPT-5.5 (medium)
272k
专有
50
$0.94
110
7.96
12.49
--
Anthropic 标志
Claude Sonnet 4.6 (max)
200k
专有
47
$1.30
67
119.47
126.99
--
Anthropic 标志
Claude Opus 4.6 (max)
1M
专有
44*
--
46
13.30
24.07
--
OpenAI 标志
GPT-5.5 (low)
272k
专有
43
$0.49
115
1.50
5.86
--
Anthropic 标志
Claude Opus 4.7 (Non-reasoning, high)
1M
专有
43*
--
59
1.49
9.91
--
Anthropic 标志
Claude Sonnet 5 (Non-reasoning)
1M
专有
42
$0.62
63
2.13
10.07
--
OpenAI 标志
GPT-5.6 Sol (Non-reasoning)
262k
专有
41
$0.30
94
1.73
7.02
--
Anthropic 标志
Claude Opus 4.5
200k
专有
41*
--
48
13.59
24.02
--
OpenAI 标志
GPT-5.4 (low)
272k
专有
39*
--
126
1.42
5.38
--
Anthropic 标志
Claude Opus 4.6 (high)
1M
专有
38*
--
46
1.83
12.79
--
SpaceXAI 标志
Grok 4.3 (high)
524k
专有
38
$0.36
198
15.09
17.61
--
Anthropic 标志
Claude 4.5 Sonnet
1M
专有
36
$0.47
52
11.61
21.21
--
SpaceXAI 标志
Grok 4.3 (medium)
524k
专有
36*
--
180
8.53
11.30
--
Anthropic 标志
Claude Sonnet 4.6 (Non-reasoning)
200k
专有
36*
--
47
1.42
12.06
--
SpaceXAI 标志
Grok 4.3 (low)
524k
专有
35*
--
161
4.19
7.30
--
OpenAI 标志
GPT-5.5 (Non-reasoning)
272k
专有
35
$0.40
110
0.80
5.33
--
Kimi 标志
Kimi K2.5
256k
开放
35
$0.26
109
1.32
33.13
27.22
Anthropic 标志
Claude Opus 4.5
200k
专有
35*
--
46
1.84
12.70
--
OpenAI 标志
GPT-5.6 Terra (Non-reasoning)
262k
专有
34
$0.20
133
0.70
4.46
--
Anthropic 标志
Claude 4.1 Opus
200k
专有
34*
--
--
--
--
--
Z AI 标志
GLM-4.7
200k
开放
34
$0.35
84
1.59
31.32
23.79
Kimi 标志
Kimi K2 Thinking
256k
开放
33*
--
122
1.44
21.89
16.36
DeepSeek 标志
DeepSeek V3.2
128k
开放
32
$0.27
103
1.38
25.73
19.48
Anthropic 标志
Claude 4.5 Haiku
200k
专有
30
$0.23
116
11.99
16.30
--
Anthropic 标志
Claude 4.5 Sonnet
1M
专有
29*
--
49
1.77
11.98
--
Anthropic 标志
Claude 4 Sonnet
1M
专有
29
$0.46
--
--
--
--
MiniMax 标志
MiniMax-M2
205k
开放
28*
--
123
1.04
21.29
16.20
Anthropic 标志
Claude 4.1 Opus
200k
专有
28*
--
--
--
--
--
OpenAI 标志
GPT-5.4 (Non-reasoning)
272k
专有
28*
--
124
0.62
4.66
--
OpenAI 标志
GPT-5.6 Luna (Non-reasoning)
262k
专有
27
$0.09
150
6.31
9.64
--
Z AI 标志
GLM-4.7
200k
开放
27*
--
89
1.40
7.05
--
Anthropic 标志
Claude 4 Sonnet
1M
专有
25*
--
--
--
--
--
SpaceXAI 标志
Grok 4.3 (Non-reasoning)
524k
专有
25
$0.91
178
0.71
3.53
--
DeepSeek 标志
DeepSeek V3.2
128k
开放
25*
--
88
1.45
7.12
--
OpenAI 标志
gpt-oss-120b (high)
131k
开放
24
$0.08
105
1.03
24.74
18.97
Anthropic 标志
Claude 4.5 Haiku
200k
专有
24*
--
102
1.03
5.94
--
Z AI 标志
GLM-4.7-Flash
200k
开放
23*
--
180
1.16
15.08
11.13
Amazon 标志
Nova 2.0 Pro Preview (medium)
256k
专有
22
$0.18
118
14.71
35.85
16.92
Alibaba 标志
Qwen3 Coder Next
128k
开放
21
$0.47
133
1.62
5.37
--
DeepSeek 标志
DeepSeek V3.1
128k
开放
21*
--
175
1.38
4.24
--
Amazon 标志
Nova 2.0 Omni (medium)
1M
专有
21*
--
--
--
--
--
Amazon 标志
Nova 2.0 Omni (medium)
1M
专有
21*
--
--
--
--
--
DeepSeek 标志
DeepSeek V3.1
128k
开放
21*
--
165
1.40
16.58
12.14
Amazon 标志
Nova 2.0 Pro Preview (low)
256k
专有
20
$0.23
123
11.68
32.00
16.25
Amazon 标志
Nova 2.0 Lite (medium)
1M
专有
19*
--
155
17.81
33.96
12.92
DeepSeek 标志
DeepSeek R1 (Jan)
128k
开放
19
$0.26
156
1.10
19.31
15.01
Amazon 标志
Nova 2.0 Lite (high)
1M
专有
18
$0.24
--
--
--
--
Amazon 标志
Nova 2.0 Lite (high)
1M
专有
18
$0.24
138
16.25
34.32
14.46
Alibaba 标志
Qwen3 235B 2507
256k
开放
18*
--
69
1.38
8.66
--
Alibaba 标志
Qwen3 Coder 480B
262k
开放
18*
--
97
1.27
6.43
--
Amazon 标志
Nova 2.0 Lite (low)
1M
专有
18*
--
156
10.57
26.55
12.78
Amazon 标志
Nova 2.0 Omni (low)
1M
专有
17*
--
--
--
--
--
Mistral 标志
Mistral Large 3
256k
开放
16
$0.08
192
1.15
3.75
--
Z AI 标志
GLM-4.7-Flash
200k
开放
16*
--
182
1.16
3.92
--
OpenAI 标志
gpt-oss-120b (low)
131k
开放
15
$0.02
102
1.05
25.67
19.70
OpenAI 标志
gpt-oss-20b (high)
131k
开放
15
$0.03
157
33.64
49.55
12.73
Amazon 标志
Nova 2.0 Pro Preview
256k
专有
14
$0.26
96
1.01
6.24
--
OpenAI 标志
gpt-oss-20b (low)
131k
开放
14*
--
101
13.01
37.74
19.78
Meta 标志
Llama 4 Maverick
128k
开放
14
$0.03
220
0.86
3.13
--
Alibaba 标志
Qwen3 Coder 30B A3B
262k
开放
14*
--
207
1.08
3.49
--
Amazon 标志
Nova Premier
1M
专有
13*
--
32
2.85
18.65
--
Amazon 标志
Nova 2.0 Lite
1M
专有
12*
--
152
1.08
4.37
--
Mistral 标志
Magistral Small 1.2
128k
开放
11
$0.29
74
1.58
35.30
26.98
Mistral 标志
Ministral 3 14B
256k
开放
11
$0.16
180
1.15
3.94
--
Amazon 标志
Nova 2.0 Omni
1M
专有
11*
--
--
--
--
--
Meta 标志
Llama 4 Scout
128k
开放
10
$0.01
155
0.83
4.06
--
Anthropic 标志
Claude 3.5 Sonnet (Oct)
200k
专有
10*
--
--
--
--
--
Meta 标志
Llama 3.3 70B
128k
开放
9
$0.11
126
1.08
5.04
--
Mistral 标志
Ministral 3 8B
256k
开放
9
$0.18
251
1.06
3.05
--
Alibaba 标志
Qwen3 32B
32.8k
开放
9*
--
158
1.18
4.34
--
Anthropic 标志
Claude 3.5 Sonnet (June)
200k
专有
8*
--
--
--
--
--
Amazon 标志
Nova Pro
300k
专有
8*
--
113
1.13
5.56
--
Meta 标志
Llama 3.1 8B
128k
开放
8
--
178
0.73
3.54
--
Google 标志
Gemma 3 27B
128k
开放
7
$0.22
68
1.43
8.78
--
NVIDIA 标志
NVIDIA Nemotron Nano 9B V2
131k
开放
7*
--
166
1.12
4.12
--
Mistral 标志
Mistral Large 2 (Jul)
128k
开放
7*
--
29
2.05
19.31
--
Amazon 标志
Nova Lite
300k
专有
7*
--
155
0.93
4.15
--
Mistral 标志
Ministral 3 3B
256k
开放
7
$0.13
408
0.98
2.20
--
Meta 标志
Llama 3.1 70B Standard
128k
开放
7*
--
97
1.32
6.46
--
Meta 标志
Llama 3.1 70B Latency Optimized
128k
开放
7*
--
123
1.27
5.34
--
Google 标志
Gemma 3 12B
128k
开放
6
$0.15
107
1.31
5.99
--
AI21 Labs 标志
Jamba 1.5 Large
256k
开放
5*
--
44
2.72
14.08
--
Anthropic 标志
Claude 3 Sonnet
200k
专有
5*
--
47
3.06
13.68
--
Amazon 标志
Nova Micro
130k
专有
5*
--
282
0.86
2.63
--
NVIDIA 标志
NVIDIA Nemotron Nano 12B v2 VL
128k
开放
5*
--
210
1.12
3.51
--
Mistral 标志
Mistral Large (Feb)
32.8k
专有
4*
--
42
1.59
13.54
--
Anthropic 标志
Claude 3 Haiku
200k
专有
4*
--
77
0.93
7.40
--
Meta 标志
Llama 3 70B
8.19k
开放
3*
--
--
--
--
--
Cohere 标志
Command-R+ (Apr)
128k
开放
3*
--
38
2.44
15.77
--
AI21 Labs 标志
Jamba 1.5 Mini
256k
开放
3*
--
53
4.47
13.99
--
Mistral 标志
Mixtral 8x7B
32.8k
开放
2*
--
24
1.10
21.97
--
Mistral 标志
Mistral 7B
8.19k
开放
2*
--
--
--
--
--
Cohere 标志
Command-R (Mar)
128k
开放
2*
--
70
1.29
8.41
--
Meta 标志
Llama 3 8B
8.19k
开放
1*
--
--
--
--
--
Google 标志
Gemma 3 4B
128k
开放
1*
--
198
1.02
3.54
--

关键定义

Maximum number of combined input & output tokens. Output tokens commonly have a significantly lower limit (varied by model).

常见问题

关于 Amazon Bedrock 的常见问题

Amazon Bedrock 提供我们跟踪的 95 个模型:Claude Opus 5 (max)Claude Opus 5 (xhigh)GPT-5.6 Sol (max)Claude Opus 5 (high)Claude Opus 5 (medium)Claude Opus 4.8 (max)GPT-5.6 Terra (max)GPT-5.5 (xhigh)Claude Opus 4.7 (max)Claude Sonnet 5 (max)GPT-5.5 (high)GPT-5.4 (xhigh)GPT-5.6 Luna (max)Claude Opus 5 (low)GPT-5.5 (medium)Claude Sonnet 4.6 (max)Claude Opus 4.6 (max)GPT-5.5 (low)Claude Opus 4.7 (Non-reasoning, high)Claude Sonnet 5 (Non-reasoning)GPT-5.6 Sol (Non-reasoning)Claude Opus 4.5GPT-5.4 (low)Claude Opus 4.6 (high)Grok 4.3 (high)Claude 4.5 SonnetGrok 4.3 (medium)Claude Sonnet 4.6 (Non-reasoning)Grok 4.3 (low)GPT-5.5 (Non-reasoning)Kimi K2.5Claude Opus 4.5GPT-5.6 Terra (Non-reasoning)GLM-4.7Kimi K2 ThinkingDeepSeek V3.2Claude 4.5 HaikuClaude 4.5 SonnetMiniMax-M2GPT-5.4 (Non-reasoning)GPT-5.6 Luna (Non-reasoning)GLM-4.7Grok 4.3 (Non-reasoning)DeepSeek V3.2gpt-oss-120b (high)Claude 4.5 HaikuGLM-4.7-FlashNova 2.0 Pro Preview (medium)Qwen3 Coder NextDeepSeek V3.1DeepSeek V3.1Nova 2.0 Pro Preview (low)Nova 2.0 Lite (medium)DeepSeek R1 (Jan)Nova 2.0 Lite (high)Qwen3 235B 2507Qwen3 Coder 480BNova 2.0 Lite (low)Mistral Large 3GLM-4.7-Flashgpt-oss-120b (low)gpt-oss-20b (high)Nova 2.0 Pro Previewgpt-oss-20b (low)Llama 4 MaverickQwen3 Coder 30B A3BNova PremierNova 2.0 LiteMagistral Small 1.2Ministral 3 14BLlama 4 ScoutLlama 3.3 70BMinistral 3 8BQwen3 32BNova ProLlama 3.1 8BGemma 3 27BNVIDIA Nemotron Nano 9B V2Mistral Large 2 (Jul)Nova LiteMinistral 3 3BLlama 3.1 70B StandardLlama 3.1 70B Latency OptimizedGemma 3 12BJamba 1.5 LargeClaude 3 SonnetNova MicroNVIDIA Nemotron Nano 12B v2 VLMistral Large (Feb)Claude 3 HaikuCommand-R+ (Apr)Jamba 1.5 MiniMixtral 8x7BCommand-R (Mar)Gemma 3 4B

Amazon Bedrock 上最智能的模型是 Claude Opus 5 (max),Intelligence Index 得分为 61。

按输出速度计算,Amazon Bedrock 上最快的模型是 Ministral 3 3B,速度为每秒 408.4 个 token。

Amazon Bedrock 上首个答案 Token 延迟最低的模型是 GPT-5.4 (Non-reasoning),延迟为 0.62 秒。延迟越低,初始响应越快。

按混合价格计算,Amazon Bedrock 上最实惠的模型是 Nova Micro,每 100 万 token 的价格为 $0.03(缓存命中/输入/输出比例为 7:2:1)。

Amazon Bedrock 上各模型价格最多相差 177 倍,从最实惠的 Nova Micro(每 100 万 token $0.03)到最昂贵的 Mistral Large (Feb)(每 100 万 token $4.80)。

Amazon Bedrock 上 95 个模型中,有 22 个支持用于结构化输出的 JSON 模式。

Amazon Bedrock 上 95 个模型中,有 93 个支持函数调用(工具使用)。

会。服务商性能可能因基础设施变化、负载均衡和更新而随时间波动。我们持续对所有服务商进行基准测试,并在“随时间变化”图表中展示历史性能趋势。

选择 Amazon Bedrock 上的模型时,请考虑:智能(适合质量敏感型任务)、输出速度(适合吞吐量密集型任务)、延迟(适合需要快速首次响应的交互式应用)、价格(适合成本敏感型工作负载),以及上下文窗口大小、JSON 模式或函数调用支持等功能。