Skip to content
VECTOR WIREAI INTELLIGENCE
⌘K
UTC
Models741Deals30d269Policy30d65, 14 pendingSources1,014

DeepSeek-V4-Flash

DeepSeekReleased Apr 24, 2026Current

Score basis

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

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.

Capable4 capabilities
  1. Long Context−11.7%1 of 3
  2. Reasoning−17.3%6 of 6
  3. Agentic−22.1%6 of 7
  4. Instruction Following−23.6%2 of 3
Limited2 capabilities
  1. Coding−25.9%10 of 10
  2. Math−48.5%5 of 5
Not rated4 capabilities

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

Price

$0.44input$1.32outputper million tokens

From Artificial Analysis · 7 providers tracked · All prices

Cheaper than 52% of 330 priced models · 3:1 input-to-output blend, log scale

Evidence

258results on85benchmarks

  • 33 independently verified
  • 110 aggregator
  • 66 vendor-reported
  • 49 cross-referenced

From 24 sources · latest Oct 8, 2026 · How verification works

API features

Tool callingStructured outputsReasoning

As listed by OpenRouter

Research

61 papers reference DeepSeek-V4-Flash

DeepSeek-V4-Flash benchmark results

258 results on 85 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

11.7% behind the leader1 of 3 ranked benchmarks measured

Show 6 more long context resultsHide 6 long context results

Reasoning

CapableFull reasoning ranking

17.3% behind the leader6 of 6 ranked benchmarks measured

Show 29 more reasoning resultsHide 29 reasoning results

Agentic

CapableFull agentic ranking

22.1% behind the leader6 of 7 ranked benchmarks measured

Show 21 more agentic resultsHide 21 agentic results

Instruction Following

CapableFull instruction following ranking

23.6% behind the leader2 of 3 ranked benchmarks measured

Show 7 more instruction following resultsHide 7 instruction following results

Coding

LimitedFull coding ranking

25.9% behind the leader10 of 10 ranked benchmarks measured

Show 29 more coding resultsHide 29 coding results

Math

LimitedFull math ranking

48.5% behind the leader5 of 5 ranked benchmarks measured

Show 11 more math resultsHide 11 math results

Factuality

Not enough dataFull factuality ranking

0 of 4 ranked benchmarks measured

Show 17 more factuality resultsHide 17 factuality results

More results

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

Show 102 more resultsHide 102 results

DeepSeek-V4-Flash: common questions

Who makes DeepSeek-V4-Flash?

DeepSeek-V4-Flash is made by DeepSeek.

When was DeepSeek-V4-Flash released?

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

What is DeepSeek-V4-Flash good at?

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

How much does DeepSeek-V4-Flash cost?

DeepSeek-V4-Flash costs $0.44 per million input tokens and $1.32 per million output tokens, according to Artificial Analysis. We track its price at 7 providers. At a mix of three input tokens to one output token, it is cheaper than 52% of the 330 priced models we track.

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

We track 258 results for DeepSeek-V4-Flash on 85 benchmarks from 24 sources, 33 of them independently verified. The latest was recorded on Oct 8, 2026.

Which API features does DeepSeek-V4-Flash support?

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

About this record

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

Tracked since
Apr 27, 2026
Newest source mention
Oct 8, 2026

Where the results come from

Verification: 258 scores · 33 independently verified · 110 aggregator-attributed · 49 vendor cross-reference · 66 vendor-reported. How these tiers are assigned

From 24 sources on 11 sites. Artificial Analysis supplies 112 of them; the 33 independently verified results come from 8 sites. Bars are coloured by trust tier.

  • artificialanalysis.ai112
  • huggingface.co94
  • livebench.ai14
  • thinkingmachines.ai13
  • api.llm-stats.com8
  • arcprize.org7
  • matharena.ai4
  • epoch.ai3
  • datasets-server.huggingface.co1
  • lmarena.ai1
  • simple-bench.com1

Also known as

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

SettingShort formLong formAPI id
lowDeepSeek V4 Flash 0731 (Low)——
highdeepseek v4 flash 0731 (high) deepseek-v4-flash high deepseek v4 flash (high)deepseek v4 flash (reasoning, high effort) deepseek v4 flash 0420 (reasoning, high effort)deepseek-v4-flash-0420-high deepseek-v4-flash-high-preview
maxdeepseek v4 flash 0731 (max) deepseek v4 flash max v4-flash max deepseek v4 flash vision (max)deepseek v4 flash (reasoning, max effort) deepseek v4 flash 0731 (reasoning, max effort) deepseek v4 flash vision (reasoning, max effort)deepseek-v4-flash (max)
Also listed asdeepseek v4 flash (non-reasoning)deepseek v4 flash (preview)deepseek v4 flash 0731deepseek-ai/deepseek-v4-flashdeepseek-ai/deepseek-v4-flash-0731deepseek-v4-flash (reasoning)deepseek-v4-flash non-thinkDeepSeek-V4-Flash Non-Think (Reasoning)deepseek-v4-flash-0420deepseek-v4-flash-latestdeepseek-v4-flash-non-reasoningdeepseek-v4-flash-thinkingunsloth/deepseek-v4-flashv4-flash non-thinkV4-Flash-0731deepseek-v4-flash-0731-ud-iq3_xxs-00001-of-00004.ggufand 14 more