CoreWeave：模型智能、性能与价格

CoreWeave

分析 CoreWeave 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 CoreWeave 提供的最佳模型。

最智能

GLM-5.2 (max)

MiniMax-M3

Kimi K2.6

Kimi K2.7 Code

GLM-5.1

Intelligence Index

共 22 个模型

速度最快

MiniMax-M3

308 t/s

Kimi K2.7 Code

287 t/s

Kimi K2.6

267 t/s

Nemotron 3 Ultra

252 t/s

Kimi K2.6

221 t/s

输出速度

共 22 个模型

价格最低

gpt-oss-20b (high)

$0.04

gpt-oss-20b (low)

$0.04

gpt-oss-120b (high)

$0.04

gpt-oss-120b (low)

$0.04

Granite 4.1 8B

$0.06

每 100 万 token 的混合价格

共 22 个模型

表示推理模型

CoreWeave 提供 22 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，CoreWeave 上表现最好的模型是 GLM-5.2 (max)（51）、MiniMax-M3（44）和Kimi K2.6（44）。
输出速度方面，最快的模型是 MiniMax-M3（308 t/s）、Kimi K2.7 Code（287 t/s）和Kimi K2.6（267 t/s）。
延迟方面，Granite 4.1 8B（0.79 秒）、Llama 3.3 70B（0.90 秒）和Qwen3 30B A3B 2507（0.92 秒）的首个答案 Token 延迟最低。
价格方面，gpt-oss-20b (high)（$0.04）、gpt-oss-20b (low)（$0.04）和gpt-oss-120b (high)（$0.04）每 100 万 token 的混合价格最低。
上下文窗口方面，DeepSeek V4 Flash (high)（1M）、DeepSeek V4 Flash（1M）和GLM-5.2 (max)（262k）支持 CoreWeave 上最大的上下文窗口。

亮点

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 v4.1 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

GDPval-AA v2

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

𝜏³-Banking

Agentic tool use

Terminal-Bench v2.1

Agentic coding & terminal use

SciCode

Coding

Humanity's Last Exam

Reasoning & knowledge

GPQA Diamond

Scientific reasoning

CritPt

Physics reasoning

AA-Omniscience Accuracy

Knowledge

AA-Omniscience Non-Hallucination Rate

1 - hallucination rate

AA-LCR

Long context reasoning

AA-Briefcase

Agentic knowledge work, Elo

AutomationBench-AA

Agentic SaaS workflows

Harvey LAB-AA

Legal agentic work, criterion pass rate

EnterpriseOps-Gym-AANew

Agentic business operations

IFBench

Instruction following

APEX-Agents-AA

Long-horizon agentic tasks

ITBench-AA

Kubernetes incident root-cause analysis

MMMU-Pro

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.

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
GLM-5.2 (max)	262k	开放	51	$0.38	189	1.42	14.64	10.58	模型服务商
MiniMax-M3	262k	开放	44	$0.11	308	1.12	9.24	6.50	模型服务商
Kimi K2.6	262k	开放	44	--	267	1.17	19.72	16.68	模型服务商
Kimi K2.7 Code	262k	开放	42	$0.28	287	1.15	10.68	7.78	模型服务商
GLM-5.1	203k	开放	40	$0.58	128	1.26	34.89	29.71	模型服务商
GLM-5 (FP8)	200k	开放	40*	--	--	--	--	--	模型服务商
Nemotron 3 Ultra	262k	开放	38	$0.16	252	1.08	12.09	9.02	模型服务商
DeepSeek V4 Flash (high)	1M	开放	37	$0.07	40	1.57	44.86	30.85	模型服务商
Kimi K2.6	262k	开放	35*	--	221	1.13	3.39	--	模型服务商
MiniMax-M2.5	197k	开放	34*	--	82	1.10	31.71	24.49	模型服务商
Gemma 4 31B	262k	开放	29	--	35	1.37	65.03	49.42	模型服务商
DeepSeek V4 Flash	1M	开放	29*	--	59	1.53	9.97	--	模型服务商
NVIDIA Nemotron 3 Super	262k	开放	25	$0.23	143	1.02	18.55	14.02	模型服务商
gpt-oss-120b (high)	131k	开放	24	$0.02	33	1.46	77.45	60.80	模型服务商
DeepSeek V3.1	128k	开放	21*	--	64	1.41	9.17	--	模型服务商
Qwen3 Coder 480B	262k	开放	18*	--	67	1.21	8.62	--	模型服务商
gpt-oss-120b (low)	131k	开放	15	$0.0042	37	1.41	68.33	53.54	模型服务商
gpt-oss-20b (high)	131k	开放	15	$0.01	121	0.89	21.49	16.48	模型服务商
gpt-oss-20b (low)	131k	开放	14*	--	93	0.94	27.69	21.40	模型服务商
Llama 3.3 70B	128k	开放	9	$0.11	77	0.90	7.38	--	模型服务商
Qwen3 30B A3B 2507	262k	开放	9*	--	147	0.92	4.33	--	模型服务商
Llama 3.1 8B	128k	开放	8	--	136	0.99	4.65	--	模型服务商
Granite 4.1 8B	131k	开放	7*	--	103	0.79	5.62	--	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 CoreWeave 的常见问题

CoreWeave 提供我们跟踪的 22 个模型：GLM-5.2 (max)、MiniMax-M3、Kimi K2.6、Kimi K2.7 Code、GLM-5.1、Nemotron 3 Ultra、DeepSeek V4 Flash (high)、Kimi K2.6、MiniMax-M2.5、Gemma 4 31B、DeepSeek V4 Flash、NVIDIA Nemotron 3 Super、gpt-oss-120b (high)、DeepSeek V3.1、Qwen3 Coder 480B、gpt-oss-120b (low)、gpt-oss-20b (high)、gpt-oss-20b (low)、Llama 3.3 70B、Qwen3 30B A3B 2507、Llama 3.1 8B和Granite 4.1 8B。

CoreWeave 上最智能的模型是 GLM-5.2 (max)，Intelligence Index 得分为 51。

按输出速度计算，CoreWeave 上最快的模型是 MiniMax-M3，速度为每秒 307.6 个 token。

CoreWeave 上首个答案 Token 延迟最低的模型是 Granite 4.1 8B，延迟为 0.79 秒。延迟越低，初始响应越快。

按混合价格计算，CoreWeave 上最实惠的模型是 gpt-oss-20b (high)，每 100 万 token 的价格为 $0.04（缓存命中/输入/输出比例为 7:2:1）。

CoreWeave 上各模型价格最多相差 43 倍，从最实惠的 gpt-oss-20b (high)（每 100 万 token $0.04）到最昂贵的 GLM-5.1（每 100 万 token $1.70）。

是，CoreWeave 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，CoreWeave 上全部 22 个模型均支持用于结构化输出的 JSON 模式。

是，CoreWeave 上全部 22 个模型均支持函数调用（工具使用）。

是，CoreWeave 提供 14 个推理模型：GLM-5.2 (max)、MiniMax-M3、Kimi K2.6、Kimi K2.7 Code、GLM-5.1、Nemotron 3 Ultra、DeepSeek V4 Flash (high)、MiniMax-M2.5、Gemma 4 31B、NVIDIA Nemotron 3 Super、gpt-oss-120b (high)、gpt-oss-120b (low)、gpt-oss-20b (high)和gpt-oss-20b (low)。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，CoreWeave 上全部 22 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 CoreWeave 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。

CoreWeave：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