Nebius:模型智能、性能与价格

Nebius
Nebius

分析 Nebius 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 Nebius 提供的最佳模型。

最智能

#1
Kimi K3 (max)
Kimi K3 (max)
57
#2
GLM-5.2 (max) (FP4)
GLM-5.2 (max) (FP4)
51
#3
MiniMax-M3 (FP8)
MiniMax-M3 (FP8)
44
#4
DeepSeek V4 Pro (max)
DeepSeek V4 Pro (max)
44
#5
Kimi K2.6
Kimi K2.6
44

Intelligence Index

共 33 个模型

速度最快

#1
Nemotron 3 Nano Omni 30B A3B Reasoning (FP8)
Nemotron 3 Nano Omni 30B A3B Reasoning (FP8)
324 t/s
#2
NVIDIA Nemotron 3 Nano
NVIDIA Nemotron 3 Nano
322 t/s
#3
Nemotron 3 Ultra
Nemotron 3 Ultra
311 t/s
#4
gpt-oss-120b (low) Base
gpt-oss-120b (low) Base
261 t/s
#5
NVIDIA Nemotron 3 Super
NVIDIA Nemotron 3 Super
243 t/s

输出速度

共 33 个模型

价格最低

#1
Nemotron 3 Nano Omni 30B A3B Reasoning (FP8)
Nemotron 3 Nano Omni 30B A3B Reasoning (FP8)
$0.08
#2
NVIDIA Nemotron 3 Nano
NVIDIA Nemotron 3 Nano
$0.08
#3
Qwen3 32B Base
Qwen3 32B Base
$0.12
#4
Qwen3 30B A3B 2507
Qwen3 30B A3B 2507
$0.12
#5
Qwen3 32B Base
Qwen3 32B Base
$0.12

每 100 万 token 的混合价格

共 33 个模型

表示推理模型

Nebius 提供 33 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,Nebius 上表现最好的模型是 Kimi K3 (max)(57)、GLM-5.2 (max) (FP4)(51)和MiniMax-M3 (FP8)(44)。
  • 输出速度方面,最快的模型是 Nemotron 3 Nano Omni 30B A3B Reasoning (FP8)(324 t/s)、NVIDIA Nemotron 3 Nano(322 t/s)和Nemotron 3 Ultra(311 t/s)。
  • 延迟方面,Qwen3 30B A3B 2507(1.12 秒)、GLM-5.2 (FP4)(1.21 秒)和Hermes 4 70B (FP8)(1.36 秒) 的首个答案 Token 延迟最低。
  • 价格方面,Nemotron 3 Nano Omni 30B A3B Reasoning (FP8)($0.08)、NVIDIA Nemotron 3 Nano($0.08)和Qwen3 32B Base($0.12) 每 100 万 token 的混合价格最低。
  • 上下文窗口方面,MiniMax-M3 (FP8)(1M)、Kimi K3 (max)(1M)和DeepSeek V4 Pro(1M) 支持 Nebius 上最大的上下文窗口。
  • Nemotron 3 Nano Omni 30B A3B Reasoning (FP8) 同时拥有最快输出和最佳价格,对吞吐量敏感且注重成本的应用很有吸引力。对于要求最高质量的任务,Kimi K3 (max) 在智能方面领先。

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
Estimate (independent evaluation forthcoming)
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
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
Kimi 标志
Kimi K3 (max)
1.05M
开放
57
$3.18
105
1.64
25.48
19.07
Z AI 标志
GLM-5.2 (max) (FP4)
432k
开放
51
$1.06
230
1.13
12.00
8.70
MiniMax 标志
MiniMax-M3 (FP8)
1.05M
开放
44
$0.45
175
1.94
16.19
11.40
DeepSeek 标志
DeepSeek V4 Pro (max)
1M
开放
44
$0.90
68
1.57
72.93
64.04
Kimi 标志
Kimi K2.6
262k
开放
44
$0.76
238
1.92
22.72
18.70
DeepSeek 标志
DeepSeek V4 Pro (high)
1M
开放
43
$0.94
54
1.44
47.85
37.10
Kimi 标志
Kimi K2.7 Code (FP4)
256k
开放
42
$0.64
87
1.76
33.01
25.52
Z AI 标志
GLM-5.1 (FP8, Base)
200k
开放
40
$0.58
37
1.81
118.78
103.34
NVIDIA 标志
Nemotron 3 Ultra
256k
开放
38
$0.59
311
2.30
11.21
7.30
Z AI 标志
GLM-5.1 (FP8, Base)
200k
开放
35*
--
36
1.84
15.86
--
Kimi 标志
Kimi K2.6
262k
开放
35*
--
218
1.92
4.21
--
Z AI 标志
GLM-5.2 (FP4)
432k
开放
34
--
204
1.21
3.65
--
Alibaba 标志
Qwen3.5 397B A17B (Base, FP4)
262k
开放
34
$0.36
120
1.99
32.79
26.62
MiniMax 标志
MiniMax-M2.5 (FP4)
196k
开放
34*
--
92
1.84
29.00
21.72
Alibaba 标志
Qwen3.5 397B A17B (Base, FP4)
262k
开放
32*
--
110
2.05
6.59
--
DeepSeek 标志
DeepSeek V4 Pro
1.05M
开放
31*
--
52
1.45
11.15
--
NVIDIA 标志
NVIDIA Nemotron 3 Super
256k
开放
25
$0.34
243
1.88
12.17
8.23
OpenAI 标志
gpt-oss-120b (high) (Base)
128k
开放
24
$0.08
221
1.03
12.35
9.06
Alibaba 标志
Qwen3 235B 2507
262k
开放
18*
--
50
1.73
11.67
--
Alibaba 标志
Qwen3 Next 80B A3B (FP8)
262k
开放
17
$0.04
102
1.09
25.65
19.65
NVIDIA 标志
Nemotron 3 Nano Omni 30B A3B Reasoning (FP8)
65.5k
开放
15*
--
324
1.01
8.73
6.18
OpenAI 标志
gpt-oss-120b (low) Base
128k
开放
15
$0.02
261
0.97
10.57
7.67
NVIDIA 标志
NVIDIA Nemotron 3 Nano
262k
开放
14
$0.03
322
1.14
8.90
6.21
Alibaba 标志
Qwen3 32B Base
32.8k
开放
12
--
31
1.99
82.32
64.27
Nous Research 标志
Hermes 4 70B (FP8)
128k
开放
10*
--
89
1.39
29.41
22.42
Meta 标志
Llama 3.3 70B Base
128k
开放
9
$0.02
14
7.08
42.58
--
NVIDIA 标志
Llama Nemotron Ultra Base
131k
开放
9*
--
52
2.34
50.08
38.19
Alibaba 标志
Qwen3 30B A3B 2507
262k
开放
9*
--
50
1.12
11.04
--
Nous Research 标志
Hermes 4 405B (FP8)
128k
开放
9*
--
32
2.63
81.09
62.77
Nous Research 标志
Hermes 4 405B (FP8)
128k
开放
9*
--
35
2.36
16.45
--
Alibaba 标志
Qwen3 32B Base
32.8k
开放
9*
--
28
1.94
19.69
--
Google 标志
Gemma 3 27B (FP8)
110k
开放
7
$0.09
59
2.08
10.49
--
Nous Research 标志
Hermes 4 70B (FP8)
128k
开放
7*
--
93
1.36
6.74
--

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 Nebius 的常见问题

Nebius 上最智能的模型是 Kimi K3 (max),Intelligence Index 得分为 57。

按输出速度计算,Nebius 上最快的模型是 Nemotron 3 Nano Omni 30B A3B Reasoning (FP8),速度为每秒 323.9 个 token。

Nebius 上首个答案 Token 延迟最低的模型是 Qwen3 30B A3B 2507,延迟为 1.12 秒。延迟越低,初始响应越快。

按混合价格计算,Nebius 上最实惠的模型是 Nemotron 3 Nano Omni 30B A3B Reasoning (FP8),每 100 万 token 的价格为 $0.08(缓存命中/输入/输出比例为 7:2:1)。

Nebius 上各模型价格最多相差 54 倍,从最实惠的 Nemotron 3 Nano Omni 30B A3B Reasoning (FP8)(每 100 万 token $0.08)到最昂贵的 Kimi K3 (max)(每 100 万 token $4.20)。

是,Nebius 提供兼容 OpenAI 的 API,可以轻松从 OpenAI 切换,或继续使用现有的 OpenAI SDK 集成。

是,Nebius 上全部 33 个模型均支持用于结构化输出的 JSON 模式。

是,Nebius 上全部 33 个模型均支持函数调用(工具使用)。

是,Nebius 上全部 33 个模型均为开放权重模型。

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 Nebius 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。