deepseek 的网站图标

DeepSeek: DeepSeek V4 Pro 0423

deepseek/deepseek-v4-pro

模型权重

DeepSeek V4 Pro is a large-scale Mixture-of-Experts model from DeepSeek with 1.6T total parameters and 49B activated parameters, supporting a 1M-token context window. It is designed for advanced reasoning, coding, and long-horizon agent workflows, with strong performance across knowledge, math, and software engineering benchmarks.

Built on the same architecture as DeepSeek V4 Flash, it introduces a hybrid attention system for efficient long-context processing. Reasoning efforts high and xhigh are supported; xhigh maps to max reasoning. It is well suited for complex workloads such as full-codebase analysis, multi-step automation, and large-scale information synthesis, where both capability and efficiency are critical

模态
输入/输出价格
50% 优惠
US$0.8719 / US$1.744每 1M
上下文
1.0M
发布时间
2026年4月24日

提供商

不同公司托管同一模型。OpenRouter 根据您选择的路由模式将请求路由到其中一家公司:Balanced(价格 + 速度)、Nitro(速度最快)或 Floor(价格最低)。

定价

客户为此模型实际支付的平均价格,与提供商公布的价格并列展示。由于缓存和折扣,实际支付的价格往往远低于标价。

加权平均值

输入
US$0.3021/百万 token
输出
US$2.074/百万 token
价格历史
当前选择无实际价格数据

性能

吞吐量是模型生成内容的速度(每秒 token 数,越高越好)。延迟是完整的往返时间(越低越好)。TTFT 是首个 token 的响应时间,即您看到内容出现前需要等待的时间(越低越好)。

吞吐量
95token/秒

P50,所有提供商中最佳

延迟
0.88秒

P50,最佳提供商

AutoExacto 基准测试
GPQA DiamondTAU-BenchParasail88.0%--Cloudflare91.9%76.1%GMICloud91.3%75.9%NovitaAI91.3%75.5%auto-routing88.4%76.9%
+13 家其他提供商

正常运行时间

正常运行时间是指过去 3 天中至少有一家提供商响应请求的时间百分比。可用性是指成功提供推理服务的时间百分比。OpenRouter 会持续监控,并在某家提供商返回错误时使用次优提供商。

正常运行时间(3 天)模型可访问。请求已路由至提供商。
100.00%
可用性(3 天)模型通过任意提供商返回了推理结果。错误和空响应均计为失败。
99.58%

过去 3 天的可用性

最近 72 小时
可用性 99.58%
3 天前2 天前昨天现在

过去 24 小时的可用性

OpenRouter Availability
99.52%
Without Routing
98.26%

当上游提供商发生错误时,如果您的请求筛选条件允许,我们可以通过路由到另一家健康的提供商来恢复服务。您可以通过以下接口以编程方式访问各提供商的正常运行时间数据: 端点 API。了解更多有关我们的 负载均衡和提供商路由选项。

基准测试

标准化评测得分,以及此模型针对共用基准测试提示词的实际输出。百分比越高越好,每项排名显示此模型在该来源所评测的模型中的位置。

基准测试得分摘要: DeepSeek: DeepSeek V4 Pro 0423 (Artificial Analysis、Design Arena 和 Vals AI)
来源基准测试得分
Artificial AnalysisDeepSeek V4 Pro 0424 (High) Coding Index58.7
Artificial AnalysisDeepSeek V4 Pro 0424 (High) GPQA Diamond90.5%
Artificial AnalysisDeepSeek V4 Pro 0424 (High) HLE35.2%
Artificial AnalysisDeepSeek V4 Pro 0424 (High) IFBench71.3%
Artificial AnalysisDeepSeek V4 Pro 0424 (High) τ²-Bench Telecom94.2%
Artificial AnalysisDeepSeek V4 Pro 0424 (High) AA-LCR70.3%
Artificial AnalysisDeepSeek V4 Pro 0424 (High) τ-Bench Banking26.2%
Artificial AnalysisDeepSeek V4 Pro 0424 (High) GDPval-AA33.0%
Artificial AnalysisDeepSeek V4 Pro 0424 (High) CritPt10.0%
Artificial AnalysisDeepSeek V4 Pro 0424 (High) Terminal-Bench Hard41.7%
Artificial AnalysisDeepSeek V4 Pro 0424 (High) Terminal-Bench 2.164.8%
Artificial AnalysisDeepSeek V4 Pro 0813 (High) Intelligence Index20.8
Artificial AnalysisDeepSeek V4 Pro 0813 (High) HLE41.5%
Artificial AnalysisDeepSeek V4 Pro 0813 (High) AA-LCR79.3%
Artificial AnalysisDeepSeek V4 Pro 0813 (High) GDPval-AA41.8%
Artificial AnalysisDeepSeek V4 Pro 0813 (High) CritPt18.6%
Artificial AnalysisDeepSeek V4 Pro 0813 (High) SciCode51.7%
Artificial AnalysisDeepSeek V4 Pro 0813 (High) AA-Omniscience Accuracy45.1%
Artificial AnalysisDeepSeek V4 Pro 0813 (High) AA-Omniscience Non-Hallucination Rate5.9%
Artificial AnalysisDeepSeek V4 Pro 0813 (Non-reasoning) Intelligence Index12.6
Artificial AnalysisDeepSeek V4 Pro 0813 (Non-reasoning) HLE10.6%
Artificial AnalysisDeepSeek V4 Pro 0813 (Non-reasoning) AA-LCR50.7%
Artificial AnalysisDeepSeek V4 Pro 0813 (Non-reasoning) GDPval-AA28.3%
Artificial AnalysisDeepSeek V4 Pro 0813 (Non-reasoning) CritPt0.0%
Artificial AnalysisDeepSeek V4 Pro 0813 (Non-reasoning) SciCode39.9%
Artificial AnalysisDeepSeek V4 Pro 0813 (Non-reasoning) Terminal-Bench 4.01.5%
Artificial AnalysisDeepSeek V4 Pro 0813 (Non-reasoning) AA-Omniscience Accuracy38.6%
Artificial AnalysisDeepSeek V4 Pro 0813 (Non-reasoning) AA-Omniscience Non-Hallucination Rate6.1%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) Coding Index59.4
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) Agentic Index26.3
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) GPQA Diamond88.8%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) HLE37.5%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) IFBench76.5%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) τ²-Bench Telecom96.2%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) AA-LCR74.7%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) τ-Bench Banking30.1%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) GDPval-AA33.3%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) CritPt12.9%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) SciCode50.8%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) Terminal-Bench Hard46.2%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) Terminal-Bench 4.014.6%
Artificial AnalysisDeepSeek V4 Pro 0424 (Max) Terminal-Bench 2.164.0%
Artificial AnalysisDeepSeek V4 Pro 0424 (Non-reasoning) GPQA Diamond71.7%
Artificial AnalysisDeepSeek V4 Pro 0424 (Non-reasoning) HLE8.2%
Artificial AnalysisDeepSeek V4 Pro 0424 (Non-reasoning) IFBench45.8%
Artificial AnalysisDeepSeek V4 Pro 0424 (Non-reasoning) τ²-Bench Telecom91.2%
Artificial AnalysisDeepSeek V4 Pro 0424 (Non-reasoning) AA-LCR53.0%
Artificial AnalysisDeepSeek V4 Pro 0424 (Non-reasoning) CritPt0.9%
Artificial AnalysisDeepSeek V4 Pro 0424 (Non-reasoning) Terminal-Bench Hard36.4%
Design ArenaDeepSeek-V4-Pro Agents Arena Full Stack Elo948
Design ArenaDeepSeek-V4-Pro Agents Arena Godotgamedev Elo1059
Design ArenaDeepSeek-V4-Pro Agents Arena Webapps Elo1000
Design ArenaDeepSeek-V4-Pro Models Arena 3D Elo1233
Design ArenaDeepSeek-V4-Pro Models Arena Asciiart Elo1131
Design ArenaDeepSeek-V4-Pro Models Arena Code Categories Elo1230
Design ArenaDeepSeek-V4-Pro Models Arena Data Visualization Elo1199
Design ArenaDeepSeek-V4-Pro Models Arena Game Development Elo1221
Design ArenaDeepSeek-V4-Pro Models Arena SVG Elo1128
Design ArenaDeepSeek-V4-Pro Models Arena UI Component Elo1210
Design ArenaDeepSeek-V4-Pro Models Arena Website Elo1227
Vals AIDeepSeek V4 CaseLaw v259.4%
Vals AIDeepSeek V4 Code Migration26.2%
Vals AIDeepSeek V4 CorpFin v261.4%
Vals AIDeepSeek V4 EMB51.6%
Vals AIDeepSeek V4 Finance Agent (v2)44.1%
Vals AIDeepSeek V4 GPQA Diamond89.4%
Vals AIDeepSeek V4 Harvey's Legal Agent Benchmark v13.8%
Vals AIDeepSeek V4 IOI v135.8%
Vals AIDeepSeek V4 LiveCodeBench87.5%
Vals AIDeepSeek V4 LegalBench80.3%
Vals AIDeepSeek V4 Legal Research Bench23.1%
Vals AIDeepSeek V4 MedCode40.5%
Vals AIDeepSeek V4 MedScribe75.1%
Vals AIDeepSeek V4 MMLU Pro87.2%
Vals AIDeepSeek V4 ProgramBench0.0%
Vals AIDeepSeek V4 ProofBench v1.116.0%
Vals AIDeepSeek V4 Public Benefits Bench v1.162.9%
Vals AIDeepSeek V4 SkillsBench51.3%
Vals AIDeepSeek V4 SWE-bench77.4%
Vals AIDeepSeek V4 Tax Agent Bench25.2%
Vals AIDeepSeek V4 TaxEval v272.1%
Vals AIDeepSeek V4 Terminal-Bench 4.011.1%
Vals AIDeepSeek V4 Vals Index38.6%
Vals AIDeepSeek V4 Vibe Code Bench v1.149.9%

应用

向此模型发送最多流量的公开应用。这能很好地反映实际生产工作负载的情况,也能提示此模型最适合哪些用例。

1.
https://nousresearch.com 的网站图标
Hermes Agent
635Btoken
2.
https://openclaw.ai/ 的网站图标
OpenClaw
126Btoken
3.
60.5Btoken
4.
https://pi.dev/ 的网站图标
pi
45.2Btoken
5.
https://janitorai.com/ 的网站图标
Janitor AI
43.8Btoken
OpenRouter 的其他部分
36.7B
前 5 大应用
17.4B

活动

此模型的 token 用量和请求流量随时间的变化。

API

可直接用于调用此模型的代码。OpenRouter 的 API 与 OpenAI 兼容——大多数 SDK 只需替换基础 URL 即可使用。切换模型时,唯一需要更改的是下方的模型标识符。

目前无法通过 EU 区域内路由使用。 区域内路由文档

文档

在 OpenRouter 上使用 DeepSeek: DeepSeek V4 Pro 0423 构建应用的指南、实用教程、博客文章和工具。

常见问题

探索更多模型