Nebius 提供哪些模型？

Nebius 提供我们跟踪的 30 个模型：GLM-5.3 (max) (FP4)、Kimi K3 (max)、GLM-5.3-Flash、DeepSeek V4 Pro 0813 (max)、DeepSeek V4 Flash 0731 (max)、GLM-5.2 (max) (FP4)、DeepSeek V4 Pro (max)、DeepSeek V4 Pro (high)、MiniMax-M3 (FP8)、Kimi K2.6、GLM-5.1 (FP8, Base)、Kimi K2.7 Code (FP4)、GLM-5.1 (Non-reasoning) (FP8, Base)、Kimi K2.6 (Non-reasoning)、Nemotron 3 Ultra、GLM-5.2 (Non-reasoning) (FP4)、Qwen3.5 397B A17B (Non-reasoning) (Base, FP4)、DeepSeek V4 Pro (Non-reasoning)、Qwen3.5 397B A17B (Base, FP4)、Nemotron 3.5 Lightning (BF16)、Nemotron 3 Super、gpt-oss-120b (high) (Base)、Qwen3 235B 2507 (Non-reasoning)、gpt-oss-120b (low) Base、Nemotron 3 Nano、Llama 3.3 70B Base、Qwen3 30B A3B 2507 (Non-reasoning)、Hermes 4 405B (FP8)、Hermes 4 405B (Non-reasoning) (FP8)和Gemma 3 27B (FP8)。

Artificial Analysis

Nebius：模型智能、性能与价格

Nebius

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

Updated

GLM-5.3 (max) (FP4)

Kimi K3 (max)

GLM-5.3-Flash

DeepSeek V4 Pro 0813 (max)

DeepSeek V4 Flash 0731 (max)

Intelligence Index

共 30 个模型

速度最快

Nemotron 3 Super

383 t/s

GLM-5.3 (max) (FP4)

377 t/s

GLM-5.2 (max) (FP4)

349 t/s

Nemotron 3 Ultra

333 t/s

MiniMax-M3 (FP8)

315 t/s

输出速度

共 30 个模型

价格最低

Nemotron 3.5 Lightning (BF16)

$0.08

Nemotron 3 Nano

$0.08

Qwen3 30B A3B 2507 (Non-reasoning)

$0.12

Gemma 3 27B (FP8)

$0.12

DeepSeek V4 Flash 0731 (max)

$0.15

每 100 万 token 的混合价格

共 30 个模型

Nebius 提供 30 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 GLM-5.3 (max) (FP4)（45）、Kimi K3 (max)（44）和GLM-5.3-Flash（42）。
输出速度方面，最快的模型是 Nemotron 3 Super（383 t/s）、GLM-5.3 (max) (FP4)（377 t/s）和GLM-5.2 (max) (FP4)（349 t/s）。
延迟方面，GLM-5.2 (Non-reasoning) (FP4)（1.04 秒）、Qwen3 30B A3B 2507 (Non-reasoning)（1.16 秒）和DeepSeek V4 Pro (Non-reasoning)（1.31 秒）的首个答案 Token 延迟最低。
价格方面，Nemotron 3.5 Lightning (BF16)（$0.08）、Nemotron 3 Nano（$0.08）和Qwen3 30B A3B 2507 (Non-reasoning)（$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 0813 (max)（1M）支持 Nebius 上最大的上下文窗口。

亮点

Updated

智能

Artificial Analysis Intelligence Index · Higher is better

速度

Output tokens per second · Higher is better

价格

USD per 1M tokens (blended) · Lower is better

智能评测

Artificial Analysis Intelligence Index

Artificial Analysis Intelligence Index v4.3 incorporates 10 evaluations: AA-Briefcase, GDPval-AA v2, AutomationBench-AA, Terminal-Bench v4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, AA-LCR v1.1

Estimate (independent evaluation forthcoming)

Artificial Analysis Intelligence Index v4.3 includes: AA-Briefcase, GDPval-AA v2, AutomationBench-AA, Terminal-Bench v4.0, 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.

Intelligence Evaluations

Intelligence evaluations measured independently by Artificial Analysis · Higher is better

AA-Briefcase

Agentic knowledge work, (Elo-500)/2000

GDPval-AA v2

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

AutomationBench-AAUpdated

Agentic SaaS workflows

Terminal-Bench v4.0New

Agentic coding & terminal use

SciCode

Coding

Humanity's Last Exam

Reasoning & knowledge

GDP.pdfNew

Professional document reasoning, All-pass

CritPt

Physics reasoning

AA-Omniscience Accuracy

Knowledge

AA-Omniscience Non-Hallucination Rate

1 - hallucination rate

AA-LCR v1.1

Long context reasoning

Harvey LAB-AA

Legal agentic work, criterion pass rate

EnterpriseOps-Gym-AA

Agentic business operations

AA-AnalystAgent

Quantitative analysis on spreadsheets & documents

IFBench

Instruction following

𝜏³-Banking

Agentic tool use

APEX-Agents-AA

Long-horizon agentic tasks

ITBench-AA

Kubernetes incident root-cause analysis

MMMU-Pro

Visual reasoning

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

Intelligence Index vs. Price

Blended at 7:2:1 (cache-input-output) · USD per 1M tokens (blended)

Most attractive quadrant

Pareto line

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

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

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

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

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

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

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
GLM-5.3 (max) (FP4)	1M	开放	45	$8.68	377	1.97	8.60	5.31	模型服务商
Kimi K3 (max)	1.05M	开放	44	$9.22	186	1.58	15.02	10.76	模型服务商
GLM-5.3-Flash	1M	开放	42	$1.19	288	0.54	9.23	6.95	模型服务商
DeepSeek V4 Pro 0813 (max)	1.05M	开放	36	$8.43	69	2.41	38.87	29.16	模型服务商
DeepSeek V4 Flash 0731 (max)	1.05M	开放	35	$0.65	307	2.09	10.23	6.51	模型服务商
GLM-5.2 (max) (FP4)	432k	开放	34	$2.94	349	1.07	8.23	5.73	模型服务商
DeepSeek V4 Pro (max)	1M	开放	31	$9.70	115	1.32	43.77	38.09	模型服务商
DeepSeek V4 Pro (high)	1M	开放	30*	--	119	1.29	22.17	16.69	模型服务商
MiniMax-M3 (FP8)	1.05M	开放	30	$1.91	315	1.86	9.80	6.35	模型服务商
Kimi K2.6	262k	开放	27	$1.91	226	2.01	23.89	19.67	模型服务商
GLM-5.1 (FP8, Base)	200k	开放	26	$2.62	34	1.86	129.14	112.45	模型服务商
Kimi K2.7 Code (FP4)	256k	开放	26	$1.89	221	1.77	14.14	10.10	模型服务商
GLM-5.1 (Non-reasoning) (FP8, Base)	200k	开放	24*	--	33	1.89	16.94	--	模型服务商
Kimi K2.6 (Non-reasoning)	262k	开放	24*	--	223	1.93	4.17	--	模型服务商
Nemotron 3 Ultra	256k	开放	23	$1.16	333	2.37	10.70	6.82	模型服务商
GLM-5.2 (Non-reasoning) (FP4)	432k	开放	22*	--	256	1.04	2.99	--	模型服务商
Qwen3.5 397B A17B (Non-reasoning) (Base, FP4)	262k	开放	21*	--	140	2.05	5.63	--	模型服务商
DeepSeek V4 Pro (Non-reasoning)	1.05M	开放	21*	--	109	1.31	5.92	--	模型服务商
Qwen3.5 397B A17B (Base, FP4)	262k	开放	19	$0.47	139	2.06	28.50	22.85	模型服务商
Nemotron 3.5 Lightning (BF16)	1.05M	开放	14	$0.09	266	1.13	10.53	7.52	模型服务商
Nemotron 3 Super	256k	开放	14	$1.64	383	1.84	8.37	5.22	模型服务商
gpt-oss-120b (high) (Base)	128k	开放	12	$0.11	302	1.03	9.31	6.62	模型服务商
Qwen3 235B 2507 (Non-reasoning)	262k	开放	12*	--	95	1.40	6.66	--	模型服务商
gpt-oss-120b (low) Base	128k	开放	10*	--	297	1.00	9.42	6.74	模型服务商
Nemotron 3 Nano	262k	开放	9	$0.02	223	1.11	12.31	8.96	模型服务商
Llama 3.3 70B Base	128k	开放	8*	--	4	15.54	135.20	--	模型服务商
Qwen3 30B A3B 2507 (Non-reasoning)	262k	开放	8*	--	42	1.16	12.99	--	模型服务商
Hermes 4 405B (FP8)	128k	开放	7*	--	43	2.67	60.54	46.30	模型服务商
Hermes 4 405B (Non-reasoning) (FP8)	128k	开放	7*	--	43	2.42	14.06	--	模型服务商
Gemma 3 27B (FP8)	110k	开放	5	$0.20	67	1.77	9.18	--	模型服务商

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 GLM-5.3 (max) (FP4)，Intelligence Index 得分为 45。

按输出速度计算，Nebius 上最快的模型是 Nemotron 3 Super，速度为每秒 383.1 个 token。

Nebius 上首个答案 Token 延迟最低的模型是 GLM-5.2 (Non-reasoning) (FP4)，延迟为 1.04 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 3.5 Lightning (BF16)，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 3.5 Lightning (BF16)（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 30 个模型均支持用于结构化输出的 JSON 模式。

是，Nebius 上全部 30 个模型均支持函数调用（工具使用）。

是，Nebius 提供 20 个推理模型：GLM-5.3 (max) (FP4)、Kimi K3 (max)、GLM-5.3-Flash、DeepSeek V4 Pro 0813 (max)、DeepSeek V4 Flash 0731 (max)、GLM-5.2 (max) (FP4)、DeepSeek V4 Pro (max)、DeepSeek V4 Pro (high)、MiniMax-M3 (FP8)、Kimi K2.6、GLM-5.1 (FP8, Base)、Kimi K2.7 Code (FP4)、Nemotron 3 Ultra、Qwen3.5 397B A17B (Base, FP4)、Nemotron 3.5 Lightning (BF16)、Nemotron 3 Super、gpt-oss-120b (high) (Base)、gpt-oss-120b (low) Base、Nemotron 3 Nano和Hermes 4 405B (FP8)。推理模型会先进行扩展思考来解决复杂问题，再给出答案。

是，Nebius 上全部 30 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。

Nebius：模型智能、性能与价格

最智能

速度最快

价格最低

智能

速度

价格