Modal：模型智能、性能与价格

Modal

本分析旨在帮助你根据使用场景，选择 Modal 提供的最佳模型。

最智能

GLM-5.3 (max)

Kimi K3 (max)

GLM-5.3-Flash

Qwen3.8 2.4T A95B

DeepSeek V4.1 Flash (max)

Intelligence Index

共 7 个模型

速度最快

Qwen3.8 2.4T A95B

221 t/s

Inkling (xhigh) (NVFP4)

202 t/s

DeepSeek V4.1 Flash (non-reasoning)

157 t/s

Kimi K3 (max)

148 t/s

GLM-5.3 (max)

136 t/s

输出速度

共 7 个模型

价格最低

DeepSeek V4.1 Flash (max)

$0.20

DeepSeek V4.1 Flash (non-reasoning)

$0.20

GLM-5.3-Flash

$0.30

GLM-5.3 (max)

$0.90

Inkling (xhigh) (NVFP4)

$0.93

每 100 万 token 的混合价格

共 7 个模型

Modal 提供 7 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，Modal 上表现最好的模型是 GLM-5.3 (max)（45）、Kimi K3 (max)（44）和GLM-5.3-Flash（42）。
输出速度方面，最快的模型是 Qwen3.8 2.4T A95B（221 t/s）、Inkling (xhigh) (NVFP4)（202 t/s）和DeepSeek V4.1 Flash (non-reasoning)（157 t/s）。各模型之间的速度差异显著，最快与最慢相差 62%。
延迟方面，DeepSeek V4.1 Flash (non-reasoning)（0.62 秒）、Qwen3.8 2.4T A95B（9.81 秒）和Inkling (xhigh) (NVFP4)（10.93 秒）的首个答案 Token 延迟最低。
价格方面，DeepSeek V4.1 Flash (max)（$0.20）、DeepSeek V4.1 Flash (non-reasoning)（$0.20）和GLM-5.3-Flash（$0.30）每 100 万 token 的混合价格最低。各模型价格最多相差 4.6 倍。
上下文窗口方面，Kimi K3 (max)（1M）、Inkling (xhigh) (NVFP4)（1M）和GLM-5.3 (max)（1M）支持 Modal 上最大的上下文窗口。

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 v4.3.2 incorporates 10 evaluations: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, AA-LCR v1.1

Artificial Analysis Intelligence Index v4.3.2 includes: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, 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.

Evaluations

Evaluations measured independently by Artificial Analysis · AA-PersonalityBench scores are percentile ranks from 1 to 100

AA-Briefcase v1.1

Agentic knowledge work, Elo

AA-Omniscience Accuracy

Knowledge

AA-Omniscience Non-Hallucination Rate

1 - hallucination rate

AutomationBench-AA

Agentic SaaS workflows

GDPval-AA v2.1

Agentic real-world work tasks, Elo

Humanity's Last Exam

Reasoning & knowledge

GDP.pdf

Professional document reasoning, All-pass

Terminal-Bench 4.0

Agentic coding & terminal use

AA-LCR v1.1

Long context reasoning

SciCodeUnder review

Coding

CritPt

Physics reasoning

Terminal-Bench-Science 0.1New

Agentic scientific research workflows in a terminal

FrontierCode 1.1New

Agentic software engineering in open-source repositories

AA-AnalystAgent

Quantitative analysis on spreadsheets & documents

Harvey LAB-AA v1.1Updated

Legal agentic work, Hallucination-Gated All-Pass Rate

EnterpriseOps-Gym-AA

Agentic business operations

AA-3DNew

Agentic interactive 3D experiences, Elo

Code2Video-AANew

Product and brand videos designed in code, Elo

ITBench-AA

Kubernetes incident root-cause analysis

MLCR-AAUpdated

Medical long context reasoning

MMMU-Pro

Visual reasoning

AA-PersonalityBench: Expressiveness ScoreNew

Rich, varied, well-crafted language

AA-PersonalityBench: Organization ScoreNew

Well-structured, scannable, answer-first

AA-PersonalityBench: Transparency ScoreNew

Truthful, acknowledges uncertainty, non-sycophantic

AA-PersonalityBench: Slop ScoreNew

Use of filler and generic language

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

Intelligence Index 与价格

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: 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 与价格

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

输出速度与价格

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

输出速度

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 or the median across providers where a first-party API is not available.

延迟

按首 Token 延迟（秒）衡量

延迟: 首个答案 Token 延迟

收到首个回答 token 的秒数 · 包含推理模型的“思考”时间

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.

缓存行为

缓存命中率

可缓存输入 token 中从缓存提供的比例 · 最近四周的中位数，更新于 2026年10月26日

The share of cacheable input tokens served from the provider's prompt cache, measured during multi-turn agentic coding sessions. We regularly measure each endpoint and report a representative rate based on recent measurements. Higher values mean more of the input is billed at the cheaper cache-hit price, where the provider offers one, which lowers the effective cost of agentic workloads. For further details, see our methodology page.

每任务成本与缓存命中率

Weighted average cost (USD) per Intelligence Index task · Cache hit rate: median of the last four weeks, updated Oct 26, 2026

Most attractive quadrant

Pareto line

Weighted average cost per Intelligence Index task. Each evaluation’s cost is calculated from input, cache hit, cache write, reasoning, and answer token prices, divided by task count, and weighted by its Intelligence Index weight.

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

端到端响应时间与价格

End-to-end response time: end-to-end seconds to output 500 tokens · USD per 1M tokens (blended)

Most attractive quadrant

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)	1M	开放	45	$1.97	136	0.73	19.06	14.66	模型服务商
Kimi K3 (max)	1.05M	开放	44	$2.73	156	1.08	17.16	12.86	模型服务商
GLM-5.3-Flash	1M	开放	42	$0.88	111	0.73	23.29	18.05	模型服务商
Qwen3.8 2.4T A95B	1M	开放	40	$1.96	214	0.74	12.42	9.34	模型服务商
DeepSeek V4.1 Flash (max)	1M	开放	39	$0.52	109	0.68	23.72	18.43	模型服务商
Inkling (xhigh) (NVFP4)	1.05M	开放	25	$0.86	203	1.02	13.35	9.86	模型服务商
DeepSeek V4.1 Flash (non-reasoning)	1M	开放	25	$0.33	147	0.63	4.02	--	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 Modal 的常见问题

Modal 提供我们跟踪的 7 个模型：GLM-5.3 (max)、Kimi K3 (max)、GLM-5.3-Flash、Qwen3.8 2.4T A95B、DeepSeek V4.1 Flash (max)、Inkling (xhigh) (NVFP4)和DeepSeek V4.1 Flash (non-reasoning)。

Modal 上最智能的模型是 GLM-5.3 (max)，Intelligence Index 得分为 45。

按输出速度计算，Modal 上最快的模型是 Qwen3.8 2.4T A95B，速度为每秒 220.5 个 token。

Modal 上首个答案 Token 延迟最低的模型是 DeepSeek V4.1 Flash (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，Modal 上最实惠的模型是 DeepSeek V4.1 Flash (max)，每 100 万 token 的价格为 $0.20（缓存命中/输入/输出比例为 7:2:1）。

Modal 上各模型价格最多相差 11 倍，从最实惠的 DeepSeek V4.1 Flash (max)（每 100 万 token $0.20）到最昂贵的 Kimi K3 (max)（每 100 万 token $2.31）。

是，Modal 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，Modal 上全部 7 个模型均支持用于结构化输出的 JSON 模式。

是，Modal 上全部 7 个模型均支持函数调用（工具使用）。

是，Modal 提供 6 个推理模型：GLM-5.3 (max)、Kimi K3 (max)、GLM-5.3-Flash、Qwen3.8 2.4T A95B、DeepSeek V4.1 Flash (max)和Inkling (xhigh) (NVFP4)。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，Modal 上全部 7 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 Modal 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。

Modal：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

智能评测

Artificial Analysis Intelligence Index

Artificial Analysis Intelligence Index

Evaluations

Intelligence Evaluation Relevance

Artificial Analysis Intelligence Index

Intelligence Index 与价格

Price-Quality Variance

上下文窗口

上下文窗口

Context Window for RAG

Context Window

价格

Intelligence Index 与价格

Price-Quality Variance

性能摘要

输出速度与价格

Output Speed

Price

速度

输出速度

Output Speed

Model Performance Representation

延迟

延迟: 首个答案 Token 延迟

Time to First Answer Token

缓存行为

缓存命中率

What Cache Hit Rate Shows

每任务成本与缓存命中率

Cost per Intelligence Index Task

端到端响应时间

端到端响应时间与价格

Price

关键定义

Context Window

常见问题

Modal 提供哪些模型？

Modal 上最智能的模型是什么？

Modal 上速度最快的模型是什么？

Modal 上哪个模型的延迟最低？

Modal 上最便宜的模型是什么？

Modal 上各模型之间的价格差异有多大？

Modal 是否兼容 OpenAI？

Modal 是否支持 JSON 模式？

Modal 是否支持函数调用？

Modal 是否提供推理模型？

Modal 是否提供开放权重模型？

Modal 的性能会随时间变化吗？

如何选择 Modal 上的模型？