Skip to content

DeepSeek · Chat

DeepSeek V4 Flash

DeepSeek 🇨🇳ChatDeepSeek familyOpen weights. One of its sources says the weights are publicly available.The maker currently lists 20 Feb 2028 as the deprecation date. Plans can change.

Newer version: DeepSeek V4.1 Flash 1017. We compared release dates and version names in our catalogue.

From Neosantara's listing:
High-speed Mixture-of-Experts (MoE) model with 284B total parameters (13B active) and 1M context window. Built for real-time interactive chats, quick reasoning tasks, web search grounding, and lightweight agentic pipelines.
llminfo IQ
129.1
Maker's price
$0.175 blended / 1M
Max context
1.1M
Listings
219

Evidence and sources

🤗 deepseek-ai/DeepSeek-V4-Flash Model architecture, as its sources state it llminfo IQ 129.1, ranked among chat models. From 229 benchmarks. Plenty of data behind it.How llminfo IQ is computedCoding IQ 107.8, 95% interval 95–120. From 18 benchmark families across 9 sources. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.How Coding IQ is computedGood value (74.6). Cheaper per task than its benchmark results predict. From 18 benchmarks with measured task costs. Plenty of data behind it.How Value for money worksYap score: tokens used per coding task vs. models with the same resultsTerse (44 ±9). Generates fewer tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate. By effort: default 44 · max 62.Median 10,210 output tokens per task at default effort.How the Yap score worksReleased 24 Apr 2026Maker HQ: CN (according to DeepSeek's own website)
All rankings: #5 on NanoGPT recently #17 on OpenRouter this month #18 on OpenRouter today #23 on OpenRouter this weekSources: Writer Pricing, Mancer data, sference data, LiteLLM data, Models.dev data, Neosantara Inference data, llminfo researchPrimary data from Writer Pricing from Mancer: architecture from sference: primary mode from LiteLLM: deprecation date from Models.dev: weights availability from Neosantara Inference: description from llminfo research: release date

Advantages and trade-offs

Compared with 10 chat models, picked mainly by popularity and IQ
We only make a claim when there's enough data on models of the same type. Updated 22 Nov, 06:07 UTC.Prices come from the model's maker where they sell it, otherwise their official partner, otherwise the market median. This comparison uses 9 maker's prices; 1 market median price.

Input & output

as stated by the model's providers
Text
Images
Text

Capabilities

Supported operations and provider-stated features

VisionReasoningFunction callingStructured outputPrompt cachingWeb search

Listings

 DeepSeekdeepseek-v4-flashSources: GenAI Prices data, LiteLLM data, Models.dev data1M$0.14cache read $0.0028$0.28——
 Orcarouterdeepseek/deepseek-v4-flash-free · freeSource: Models.dev data1M$0.00$0.00——
 Unorouterdeepseek-v4-flash:free · freeSource: Models.dev data1M$0.00$0.00——
 NanoGPTInference by CoralbricksRegions: UNKdeepseek/deepseek-v4-flash-latest · latest, standardSource: NanoGPTNot published by the provider$0.0053cache read $0.0032$0.5985131.62,837ms
 Hugging FaceInference by Relacedeepseek-ai/DeepSeek-V4-FlashSource: Hugging Face Inference Providers data1M$0.0138$0.45——
 OpenRouterInference by Relacedeepseek/deepseek-v4-flash · fp4Source: OpenRouter data docs1M$0.0138$0.45139405ms
 Kilo~deepseek/deepseek-v4-flash-latest · latestSource: Kilo AI Gateway data1M$0.0443cache read $0.035$0.30——
 Adaptiondeepseek-v4-flashSource: Models.dev data1M$0.0444cache read $0.0014$0.1333——
 Fastrouterdeepseek/deepseek-v4-flashSource: FastRouter data1M$0.06cache read $0.015$0.18——
 Llmgatewaydeepseek-v4-flashSource: LLMGateway data1.1M$0.076cache read $0.014$0.153——

18 listings without a price are hidden.

Included in plans

MiniStepFunSubscription$6.99/moPlan page
PlusStepFunSubscription$9.99/moPlan page
ProStepFunSubscription$29/moPlan page
MaxStepFunSubscription$99/moPlan page

Benchmarks

Named effort levels run from lowest to highest. Select a result to see its source details.

Human preference

Arena ELO50 related benchmarks · 50 results
Source: LMArena data · Nov 2026
BenchmarkScore
Arena ELO
Chinese
Creative Writing
English
Show 46 more benchmarksHide 46 benchmarks
BenchmarkScore
Exclude Ties
Expert
French
German
Hard Prompts
Hard Prompts English
Industry Business and Management and Financial Operations
Industry Entertainment and Sports and Media
Industry Life and Physical and Social Science
Industry Mathematical
Industry Medicine and Healthcare
Industry Software and It Services
Industry Writing and Literature and Language
Japanese
Korean
Longer Query
Multi Turn
Non English
Polish
Russian
Spanish
Style-Controlled
Text Factuality
Text Factuality - Chinese
Text Factuality - Creative Writing
Text Factuality - English
Text Factuality - Exclude Ties
Text Factuality - Expert
Text Factuality - French
Text Factuality - Hard Prompts
Text Factuality - Hard Prompts English
Text Factuality - Industry Business and Management and Financial Operations
Text Factuality - Industry Entertainment and Sports and Media
Text Factuality - Industry Life and Physical and Social Science
Text Factuality - Industry Mathematical
Text Factuality - Industry Medicine and Healthcare
Text Factuality - Industry Software and It Services
Text Factuality - Industry Writing and Literature and Language
Text Factuality - Japanese
Text Factuality - Longer Query
Text Factuality - Multi Turn
Text Factuality - Non English
Text Factuality - Russian
Text Factuality - Spanish
Design Arena EloSource: Design Arena Leaderboards data · Nov 2026

Agentic

BrowseComp3 results
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Source: CyberGym leaderboard data docs
By agent · trials: 1 · features: dynamic
By agent · trials: 1 · features: dynamic, test-time mem.
agent: FangcunCyber, trials: 1, features: test-time mem.
agent: Gusion, trials: 1
agent: Sangfor AI, trials: 1, features: Orchestration, Multi-stage, dynamic
GAIA4 results
Source: GAIA Leaderboard data · Aug 2026
Difficulty level
Difficulty not reported
GDPval-AADeepSeek V4 official model card · Think Max · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
MCP Atlas4 results
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SWE-Bench Pro4 results
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: Hugging Face Official Benchmarks data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Source: OpenRouter benchmark exports
23 Apr 2026 version · 90-day results
Across 22 providers lowest 70% median 75.1% highest 81.7%OpenRouter's automatic routing 72.3%
Show results from 22 providers and OpenRouter's automatic routingHide results from 22 providers and OpenRouter's automatic routing
  • OpenInference
  • Reka
  • DeepInfra
  • Cloudflare
  • GMICloud
  • Parasail
  • Relace
  • Wafer
  • DigitalOcean
  • Venice
  • Azure
  • AtlasCloud
  • SiliconFlow
  • Baidu
  • Alibaba
  • StreamLake
  • NextBit
  • Crusoe
  • CoreWeave
  • OpenRouter's automatic routing
  • Novita
  • Mancer 2
  • Phala

Coding

Arena ELOSource: LMArena data · Nov 2026
Codeforces RatingDeepSeek V4 official model card · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
LiveCodeBench4 results
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SkillsBench2 results
Reported by Dots Studio / Xiaohongshu · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Dots Studio official appendix · unstarred comparison cell · version, task count, harness and reasoning setting not independently specified
Source: Hugging Face Official Benchmarks data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SWE-bench Verified4 related benchmarks · 10 results
SWE-bench Verified
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: Hugging Face Official Benchmarks data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SWE-bench Multilingual
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: Hugging Face Official Benchmarks data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SWE-rebench 2026-03
Source: SWE-rebench
SWE-rebench 2026-04
Source: SWE-rebench

Science

GPQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
GPQA Diamond27 results
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: OpenRouter benchmark exports
23 Apr 2026 version · 90-day results
Across 23 providers lowest 34.7% median 87% highest 90.1%OpenRouter's automatic routing 86.1%
Show results from 23 providers and OpenRouter's automatic routingHide results from 23 providers and OpenRouter's automatic routing
  • Crusoe
  • SiliconFlow
  • Cloudflare
  • GMICloud
  • Novita
  • Alibaba
  • DigitalOcean
  • Azure
  • Phala
  • Wafer
  • CoreWeave
  • Parasail
  • DeepInfra
  • NextBit
  • Relace
  • Baidu
  • OpenInference
  • Mancer 2
  • OpenRouter's automatic routing
  • Fireworks
  • Venice
  • AtlasCloud
  • StreamLake
  • Reka

General

Reasoning

ApexDeepSeek V4 official model card · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
MRCR 1MSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Toolathlon VerifiedDeepSeek V4 official model card · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.

Multilingual

EKI EstonianSource: Keelemudelite mõõdupuu data docs

Long context

Math

HMMT 2026 FebDeepSeek V4 official model card · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Factuality

Chinese SimpleQADeepSeek V4 official model card · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
CSimpleQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
SimpleQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
SimpleQA VerifiedDeepSeek V4 official model card · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.

Knowledge

MMLU-Pro4 results
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4 official model card
Source: Hugging Face Official Benchmarks data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Safety

PhareSource: Phare

Instruction following

Vision

Other

Apex ShortlistDeepSeek V4 official model card · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Claw-Eval v1.1Reported by Claw-Eval · via BenchBoard docs
MRCR · 1M · MMRDeepSeek V4 official model card · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Nejumi 428 related benchmarks · 28 results
ALT_バイアス
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
ALT_倫理・道徳
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
ALT_制御性
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
ALT_堅牢性
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
ALT_毒性
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
ALT_真実性
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
Show 22 more resultsHide 22 results
AVG_jaster_0shot
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
AVG_jaster_2shots
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_アプリケーション開発
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_コーディング
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_一般的知識
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_基礎的言語性能
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_専門的知識
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_応用的言語性能
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_情報検索
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_意味解析
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_抽象的推論
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_推論能力
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_数学的推論
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_構文解析
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_知識・質問応答
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_翻訳
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_表現
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_論理的推論
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
GLP_関数呼び出し
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
TOTAL_SCORE
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
アラインメント(ALT)_AVG
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max
汎用的言語性能(GLP)_AVG
Source: Nejumi Leaderboard 4 data docs
deepseek-ai/DeepSeek-V4-Flash: reasoning-max

Data credits: SEA-HELM leaderboard, AI Singapore (AI Products team), National University of Singapore. Published aggregate scores; results licence not stated (SEA-HELM code is MIT). · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · LMArena, CC BY 4.0 (licence) · Baromètre de l’écriture IA, Johannes Eva; public numeric ranking data, © 2026 Johannes Eva, no results licence stated · Data by LLM Stats · BullshitBench by Peter Gostev, MIT (licence) · Hugging Face official benchmark leaderboards; licence varies by benchmark · EQ-Bench (Samuel J. Paech) — cite paech2023eqbench · CyberGym leaderboard (Wang, Shi, He, Cai, Zhang, Song; UC Berkeley), ICLR 2026. Results evaluated and submitted by the listed teams. No licence is stated for the result data; the repository's Apache-2.0 licence covers its code only. · Keelemudelite mõõdupuu: Independent LLM leaderboard for Estonian. Eesti Keele Instituut. https://moodupuu.eki.ee, CC BY 4.0. Latest per-benchmark results selected with partial-run flags retained. (licence) · OpenRouter public benchmark leaderboard and search-lane exports, CC BY 4.0. Fractional source scores converted to percentages; immutable dated snapshots. (licence) · Nejumi Leaderboard 4, Weights & Biases Japan; public publisher-selected numeric evaluation results · Phare (Giskard), CC BY 4.0 (licence) · SnakeBench (Greg Kamradt / ARC Prize Foundation), MIT (licence) · UzLiB, Tahrirchi (Abror Shopulatov). Public LEADERBOARD.md numbers; repository MIT covers software and documentation, no separate results licence stated and none asserted. · Clinical Benchmarks (clinicalbenchmarks.ai), CC BY 4.0; fractions converted to percentages; source documents retain their own terms (licence)

Also known as: aihubmix/deepseek-v4-flash, azure_ai/deepseek-v4-flash, azure_ai/fw-deepseek-v4-flash, cline-pass/deepseek-v4-flash, dashscope/deepseek-v4-flash, deepinfra/deepseek-ai/deepseek-v4-flash, deepinfra/deepseek-v4-flash, deepinfra/deepseek-v4-flash-0424, deepseek v4 flash, deepseek-4-flash, deepseek-ai/deepseek-v4-flash, deepseek-v4-flash-0423, deepseek-v4-flash-260425, deepseek-v4-flash-free, deepseek-v4-flash-high, deepseek-v4-flash-latest, deepseek-v4-flash:free, deepseek-v4-flash:web, deepseek/deepseek-v4-flash, deepseek/deepseek-v4-flash-0423, deepseek/deepseek-v4-flash-0423:thinking, deepseek/deepseek-v4-flash-202605, deepseek/deepseek-v4-flash-fast, deepseek/deepseek-v4-flash-free, deepseek/deepseek-v4-flash-latest, deepseek/deepseek-v4-flash:thinking, e2ee-deepseek-v4-flash, fireworks_ai/accounts/fireworks/models/deepseek-v4-flash, fireworks_ai/deepseek-v4-flash, flexai/deepseek-v4-flash, gmi/deepseek-ai/deepseek-v4-flash, libertai/deepseek-v4-flash, nebius/deepseek-ai/deepseek-v4-flash, novita/deepseek/deepseek-v4-flash, novita/deepseek/deepseek-v4-flash-0424, openrouter/deepseek/deepseek-v4-flash, parasail/deepseek-v4-flash-0424, pinstripes/ps/deepseek-v4-flash, qwen_ai_platform/deepseek-v4-flash, qwencloud/deepseek-v4-flash, runware/deepseek-v4-flash, streamlake/deepseek-v4-flash, tencent/deepseek-v4-flash, tensormesh/deepseek-ai/deepseek-v4-flash, wandb/deepseek-ai/deepseek-v4-flash

Like this one. Tap or hover any dotted value, label or icon to see where it comes from and how it is measured. open details when tapped or hovered. How this data works