Skip to content
VECTOR WIREAI INTELLIGENCE
⌘K
UTC
Models741Deals30d271Policy30d67, 15 pendingSources1,014

DeepSeek-V4-Pro

DeepSeekReleased Apr 24, 2026Current

Score basis

DeepSeek-V4-Pro is capable in long context, reasoning, and agentic tasks; and behind the leaders in coding, instruction following, factuality, and math. Too few results yet to rate safety, multimodal tasks, or multilingual tasks.

Capability profile

Bars show the model's median result as a share of the leading model's, per capability. Select a row to see its results.

Capable3 capabilities
  1. Long Context−19.1%1 of 3
  2. Reasoning−22.1%5 of 6
  3. Agentic−24.2%6 of 7
Limited4 capabilities
  1. Coding−25.5%10 of 10
  2. Instruction Following−30.5%2 of 3
  3. Factuality−33.2%4 of 4
  4. Math−51.2%3 of 5
Not rated3 capabilities

Too few results yet to rate Safety, Multimodal or Multilingual.

Price

$0.43input$0.87outputper million tokens

From Artificial Analysis · 5 providers tracked · All prices

Cheaper than 55% of 330 priced models · 3:1 input-to-output blend, log scale

Evidence

256results on111benchmarks

  • 27 independently verified
  • 84 aggregator
  • 86 vendor-reported
  • 59 cross-referenced

From 23 sources · latest Oct 8, 2026 · How verification works

API features

Tool callingStructured outputsReasoning

As listed by OpenRouter

Research

41 papers reference DeepSeek-V4-Pro

DeepSeek-V4-Pro benchmark results

256 results on 111 benchmarks, grouped by capability. Each bar is the result as a share of the capability leader's; every result links to its source.

Long Context

CapableFull long context ranking

19.1% behind the leader1 of 3 ranked benchmarks measured

Show 4 more long context resultsHide 4 long context results

Reasoning

CapableFull reasoning ranking

22.1% behind the leader5 of 6 ranked benchmarks measured

Show 23 more reasoning resultsHide 23 reasoning results

Agentic

CapableFull agentic ranking

24.2% behind the leader6 of 7 ranked benchmarks measured

Show 20 more agentic resultsHide 20 agentic results

Coding

LimitedFull coding ranking

25.5% behind the leader10 of 10 ranked benchmarks measured

Show 24 more coding resultsHide 24 coding results

Instruction Following

LimitedFull instruction following ranking

30.5% behind the leader2 of 3 ranked benchmarks measured

Show 6 more instruction following resultsHide 6 instruction following results

Factuality

LimitedFull factuality ranking

33.2% behind the leader4 of 4 ranked benchmarks measured

Show 12 more factuality resultsHide 12 factuality results

Math

LimitedFull math ranking

51.2% behind the leader3 of 5 ranked benchmarks measured

Show 13 more math resultsHide 13 math results

More results

Benchmarks outside the capability baskets. They are not ranked against a leader.

Show 118 more resultsHide 118 results

DeepSeek-V4-Pro: common questions

Who makes DeepSeek-V4-Pro?

DeepSeek-V4-Pro is made by DeepSeek.

When was DeepSeek-V4-Pro released?

DeepSeek-V4-Pro was released on Apr 24, 2026, according to Artificial Analysis.

What is DeepSeek-V4-Pro good at?

DeepSeek-V4-Pro is capable in long context, reasoning, and agentic tasks; and behind the leaders in coding, instruction following, factuality, and math. Too few results yet to rate safety, multimodal tasks, or multilingual tasks.

How much does DeepSeek-V4-Pro cost?

DeepSeek-V4-Pro costs $0.43 per million input tokens and $0.87 per million output tokens, according to Artificial Analysis. We track its price at 5 providers. At a mix of three input tokens to one output token, it is cheaper than 55% of the 330 priced models we track.

How many benchmarks has DeepSeek-V4-Pro been tested on?

We track 256 results for DeepSeek-V4-Pro on 111 benchmarks from 23 sources, 27 of them independently verified. The latest was recorded on Oct 8, 2026.

Which API features does DeepSeek-V4-Pro support?

OpenRouter lists tool calling, structured outputs, and reasoning for DeepSeek-V4-Pro.

About this record

Where DeepSeek-V4-Pro's numbers come from, and every name it appears under.

Tracked since
Jun 27, 2026
Newest source mention
Sep 30, 2026

Where the results come from

Verification: 256 scores · 27 independently verified · 84 aggregator-attributed · 59 vendor cross-reference · 86 vendor-reported. How these tiers are assigned

From 23 sources on 11 sites. Hugging Face supplies 126 of them; the 27 independently verified results come from 8 sites. Bars are coloured by trust tier.

  • huggingface.co126
  • artificialanalysis.ai86
  • api.llm-stats.com16
  • livebench.ai7
  • datasets-server.huggingface.co4
  • matharena.ai4
  • raw.githubusercontent.com4
  • epoch.ai3
  • thinkingmachines.ai3
  • lmarena.ai2
  • simple-bench.com1

Also known as

How our sources name DeepSeek-V4-Pro at each reasoning setting.

SettingShort formLong formAPI id
highdeepseek v4 pro (high)deepseek v4 pro (reasoning, high effort)deepseek-v4-pro-high deepseek-v4-pro-high-preview deepseek-v4-pro-high-20260813
maxdeepseek v4 pro (max) deepseek v4 pro max ds-v4-pro maxdeepseek v4 pro (reasoning, max effort)—
Also listed asdeep seek v4 prodeepseek v4 pro (non-reasoning)deepseek v4 pro previewdeepseek-ai/deepseek-v4-prodeepseek-v4-pro (preview)deepseek-v4-pro-thinkingdeepseek's v4 prodeepseekv4prods v4 proV4-Pro Non-Thinkdeepseek v4proDS-v4-Pro-1.6T-A49Bdsv4-prodeepseek-v4-pro-non-reasoning