Skip to main content

Model profile

DeepSeek V4 Flash (Max)

DeepSeekCurrentReleased Apr 24, 2026
Data verified
Overall Score
Unranked
Arena Elo
Not listed
Eligible category ranks
3of 8
Price (1M tokens)
$0.14 in / $0.28 out
API pricing
Speed
Not listed
Context
1M

Evidence coverage

45 of 323 tracked benchmarks are published. 23 are verified and 22 provisional. 7 of 8 categories are measured.

Updated July 23, 2026Methodology
Published / tracked
45 / 323
Verified
23
Provisional
22
Categories with evidence
7 / 8

Evidence by category

  • Agentic15 benchmarks
    Mixed evidence
  • Coding7 benchmarks
    Mixed evidence
  • Reasoning4 benchmarks
    Mixed evidence
  • Knowledge13 benchmarks
    Mixed evidence
  • Math4 benchmarks
    Verified
  • Multilingual0 benchmarks
    Not measured
  • Multimodal1 benchmark
    Reported
  • Inst. Following1 benchmark
    Reported
Open WeightSelf-hostReasoning
Confidence:
Medium
flash-reasoning

BenchLM is tracking DeepSeek V4 Flash (Max), but this profile is currently excluded from the public leaderboard because it still lacks enough non-generated benchmark coverage to rank safely. Only non-generated public benchmark rows appear below.

DeepSeek V4 Flash (Max) is a open weight model with a 1M token context window. It uses explicit chain-of-thought reasoning, which typically improves performance on math and complex reasoning tasks at the cost of higher latency and token usage.

DeepSeek V4 Flash (Max) sits inside the DeepSeek V4 family alongside DeepSeek V4 Pro (Max), DeepSeek V4 Pro (High), DeepSeek V4 Flash (High), DeepSeek V4 Pro, DeepSeek V4 Pro Base, DeepSeek V4 Flash Base, DeepSeek V4 Flash. This profile currently has 45 of 323 tracked benchmarks. BenchLM only exposes non-generated benchmark rows publicly, so missing categories stay blank until a sourced evaluation is available.

Its strongest category is Knowledge (#39), while its weakest is Agentic (#52). This performance profile makes it particularly effective for knowledge-intensive tasks like research, analysis, and factual Q&A.

Peer position

Exact provisional scores and ranks for the closest listed peers. A score can appear before a model clears the evidence threshold for a rank, so equal scores can have different rank states.

Range 77.4483.93

  1. Claude Mythos 5
    Anthropic
    #183.93
    Claude Mythos 5 is #1 with a score of 83.93.
    Compare
  2. Claude Fable 5
    Anthropic
    #283.68
    Claude Fable 5 is #2 with a score of 83.68.
    Compare
  3. GPT-5.6 Sol
    OpenAI
    #381.96
    GPT-5.6 Sol is #3 with a score of 81.96.
    Compare
  4. Kimi K3
    Moonshot AI
    #480.96
    Kimi K3 is #4 with a score of 80.96.
    Compare
  5. Claude Opus 4.8
    Anthropic
    #578.34
    Claude Opus 4.8 is #5 with a score of 78.34.
    Compare
  6. Muse Spark 1.1
    Meta
    #677.44
    Muse Spark 1.1 is #6 with a score of 77.44.
    Compare
  7. DeepSeek V4 Flash (Max)Current model
    DeepSeek
    UnrankedNot measured
    DeepSeek V4 Flash (Max) is Unranked with a score of Not measured.

Category percentile

More

Relative position among models eligible for each sourced category. A higher percentile means a stronger position within that category's ranked cohort; 100 is highest.

  1. Knowledge25%
    Eligible cohort rank #39 of 52Category score 61.6
  2. Coding60%
    Eligible cohort rank #50 of 122Category score 52.0
  3. Agentic57%
    Eligible cohort rank #52 of 119Category score 49.5

Category evidence

Scores and ranks appear only where this model has published benchmark evidence. Categories without displayable source records remain not measured.

Category scores, ranks, weighting, benchmark coverage, and evidence status
CategoryScore
AgenticRank #52 of 119Percentile 57thWeight 22%15 benchmarksMixed sources49.5
CodingRank #50 of 122Percentile 60thWeight 20%7 benchmarksMixed sources52.0
ReasoningWeight 17%4 benchmarksMixed sourcesScore pending
KnowledgeRank #39 of 52Percentile 25thWeight 12%13 benchmarksMixed sources61.6
MathRank Not rankedWeight 5%4 benchmarksVerified81.6
MultilingualWeight 7%0 benchmarksNot measuredNot measured
MultimodalWeight 12%1 benchmarkReportedScore pending
Inst. FollowingWeight 5%1 benchmarkReportedScore pending

Benchmark Details

Rows below have a displayable published verification record. Each source link and provenance note remains in the page HTML while its category is closed. Source-unverified manual rows and generated rows stay hidden.

Agentic15 benchmarks
Terminal-Bench 2.0Provider exact
56.9%Weighted 38%
Source: DeepSeek-V4 technical reportProvenance: Provider exact
BrowseCompProvider exact
73.2%Weighted 28%
Source: DeepSeek-V4 technical reportProvenance: Provider exact
HLE w/ toolsProvider exact

Humanity's Last Exam with tools

45.1%Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
MCP AtlasProvider exact
69%Display only
Source: DeepSeek-V4 technical reportProvenance: DeepSeek labels this row MCPAtlas Public in Table 7; BenchLM stores it on the existing mcpAtlas key.
GDPval-AAProvider exact
1189Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
ToolathlonProvider exact
47.8%Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
AA Agentic IndexReported

Artificial Analysis Agentic Index

31.1%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
τ²-bench resultsReported

τ²-Bench Tool-Agent-User Evaluation

95%Display only
Source: Artificial Analysis: tau2-bench leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
GDPval-AAReported

GDPval-AA normalized

34.4%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA BriefcaseReported

Artificial Analysis Briefcase

831Display only
Source: Artificial Analysis: aa-briefcase leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA EnterpriseOps-GymReported

Artificial Analysis EnterpriseOps-Gym

39.6%Display only
Source: Artificial Analysis: enterprise-ops-gym-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA Harvey LABReported

Artificial Analysis Harvey LAB-AA

81.3%Display only
Source: Artificial Analysis: harvey-lab-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA ITBenchReported

Artificial Analysis ITBench-AA

31.5%Display only
Source: Artificial Analysis: itbench-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA Tau3 BankingReported

Artificial Analysis Tau3-Banking

22.9%Display only
Source: Artificial Analysis: tau3-banking leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
terminalBenchHardReported
35.6%Display only
Source: Artificial Analysis: terminalbench-hard leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
Coding7 benchmarks
SWE-bench VerifiedProvider exact

Software Engineering Benchmark Verified

79%Weighted 16%
Source: DeepSeek-V4 technical reportProvenance: Provider exact
SWE-bench ProProvider exact
52.6%Weighted 10%
Source: DeepSeek-V4 technical reportProvenance: Provider exact
CodeforcesProvider exact

Codeforces Rating

3052.0Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
SWE MultilingualProvider exact
73.3%Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
Terminal-Bench 2.0Provider exact
56.9%Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
AA Coding IndexReported

Artificial Analysis Coding Index

56.2%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA-SciCodeReported

Artificial Analysis SciCode

44.9%Display only
Source: Artificial Analysis: scicode leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
Reasoning4 benchmarks
MRCR 1MProvider exact
78.7%Display only
Source: DeepSeek-V4 technical reportProvenance: DeepSeek reports MRCR 1M in Table 7. BenchLM stores this as a display-only long-context row distinct from existing MRCRv2 keys.
CorpusQA 1MProvider exact
60.5%Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
AA-LCRReported

Artificial Analysis Long Context Reasoning

63.0%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
CritPtReported

Critical Physics Tasks

7.1%Display only
Source: Artificial Analysis: critpt leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
Knowledge13 benchmarks
HLEProvider exact

Humanity's Last Exam

34.8%Weighted 45%
Source: DeepSeek-V4 technical reportProvenance: Provider exact
MMLU-ProProvider exact

Massive Multitask Language Understanding Professional

86.2%Weighted 30%
Source: DeepSeek-V4 technical reportProvenance: Provider exact
SimpleQAProvider exact

Measuring Short-Form Factuality in Large Language Models

34.1%Weighted 11%
Source: DeepSeek-V4 technical reportProvenance: Provider exact
GPQAProvider exact

Graduate-Level Google-Proof Q&A

88.1%Weighted 7%
Source: DeepSeek-V4 technical reportProvenance: DeepSeek reports GPQA Diamond in Tables 1/7. BenchLM maps Table 7 GPQA Diamond values into both gpqa and gpqaDiamond for comparison consistency.
Chinese-SimpleQAProvider exact
78.9%Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
GPQA-DProvider exact

GPQA Diamond

88.1%Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
Artificial Analysis Intelligence IndexReported
40.3%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA-GPQA DiamondReported

Artificial Analysis GPQA Diamond

89.4%Display only
Source: Artificial Analysis: gpqa-diamond leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA-HLEReported

Artificial Analysis Humanity's Last Exam

32.1%Display only
Source: Artificial Analysis: humanitys-last-exam leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA-Omniscience IndexReported

Artificial Analysis Omniscience Index

-22.9%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA-Omniscience AccuracyReported

Artificial Analysis Omniscience Accuracy

37.2%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA-Omniscience Hallucination RateReported

Artificial Analysis Omniscience Hallucination Rate

95.8%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA Openness IndexReported

Artificial Analysis Openness Index

50.0%Display only
Source: Artificial Analysis: artificial-analysis-openness-index leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
Math4 benchmarks
HMMT Feb 2026Provider exact

Harvard-MIT Mathematics Tournament February 2026

94.8%Weighted 25%
Source: DeepSeek-V4 technical reportProvenance: Provider exact
IMOAnswerBenchProvider exact
88.4%Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
ApexProvider exact
33.0%Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
Apex ShortlistProvider exact
85.7%Display only
Source: DeepSeek-V4 technical reportProvenance: Provider exact
Multimodal1 benchmark
Design Arena WebsiteReported

Design Arena Website Elo

1238Display only
Source: OpenRouter model benchmarksProvenance: Display-only Design Arena Website Elo synced from OpenRouter model benchmark metadata. It is excluded from BenchLM weighted scoring.
Inst. Following1 benchmark
AA-IFBenchReported

Artificial Analysis IFBench

79.2%Display only
Source: Artificial Analysis: ifbench leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.

Frequently Asked Questions

How does DeepSeek V4 Flash (Max) perform overall in AI benchmarks?

DeepSeek V4 Flash (Max) has 45 published benchmark scores on BenchLM, but it does not yet have enough non-generated coverage to receive a global overall rank.

Is DeepSeek V4 Flash (Max) good for knowledge and understanding?

DeepSeek V4 Flash (Max) ranks #39 out of 52 models in knowledge and understanding benchmarks with an average score of 61.6. There are stronger options in this category.

Is DeepSeek V4 Flash (Max) good for coding and programming?

DeepSeek V4 Flash (Max) ranks #50 out of 122 models in coding and programming benchmarks with an average score of 52. There are stronger options in this category.

Is DeepSeek V4 Flash (Max) good for mathematics?

DeepSeek V4 Flash (Max) has visible benchmark coverage in mathematics, but BenchLM does not currently assign it a global category rank there.

Is DeepSeek V4 Flash (Max) good for reasoning and logic?

DeepSeek V4 Flash (Max) has visible benchmark coverage in reasoning and logic, but BenchLM does not currently assign it a global category rank there.

Is DeepSeek V4 Flash (Max) good for agentic tool use and computer tasks?

DeepSeek V4 Flash (Max) ranks #52 out of 119 models in agentic tool use and computer tasks benchmarks with an average score of 49.5. There are stronger options in this category.

Is DeepSeek V4 Flash (Max) good for multimodal and grounded tasks?

DeepSeek V4 Flash (Max) has visible benchmark coverage in multimodal and grounded tasks, but BenchLM does not currently assign it a global category rank there.

Is DeepSeek V4 Flash (Max) good for instruction following?

DeepSeek V4 Flash (Max) has visible benchmark coverage in instruction following, but BenchLM does not currently assign it a global category rank there.

Is DeepSeek V4 Flash (Max) open source?

Yes, DeepSeek V4 Flash (Max) is an open weight model created by DeepSeek, meaning it can be downloaded and run locally or fine-tuned for specific use cases.

Which sibling models are related to DeepSeek V4 Flash (Max)?

DeepSeek V4 Flash (Max) belongs to the DeepSeek V4 family. Related variants on BenchLM include DeepSeek V4 Pro (Max), DeepSeek V4 Pro (High), DeepSeek V4 Flash (High), DeepSeek V4 Pro, DeepSeek V4 Pro Base, DeepSeek V4 Flash Base, DeepSeek V4 Flash.

Does DeepSeek V4 Flash (Max) have full benchmark coverage on BenchLM?

Not yet. DeepSeek V4 Flash (Max) currently has 45 published benchmark scores out of the 323 benchmarks BenchLM tracks. BenchLM only exposes non-generated public benchmark rows, so missing categories stay blank until a sourced evaluation is available.

What is the context window size of DeepSeek V4 Flash (Max)?

DeepSeek V4 Flash (Max) has a published context window of 1M, which determines how much text it can process in a single interaction.

Last updated: July 23, 2026 · Runtime metrics stay blank until BenchLM has a sourced snapshot.

Choose with this week’s evidence

Join 2,000+ readers for ranking moves, new releases, pricing changes, and the evidence behind them.

Free. One email per week.