Skip to main content

Model profile

DeepSeek V3

DeepSeekEstablishedReleased Dec 26, 2024
Data verified
Overall Score
44.97Public #147 of 200Verified #72 of 99
Arena Elo
1358
Eligible category ranks
2of 8
Price (1M tokens)
$0.27 in / $1.1 out
API pricing
Speed
Not listed
Context
128K

Evidence coverage

22 of 323 tracked benchmarks are published. 1 is verified and 21 provisional. 7 of 8 categories are measured.

Updated July 21, 2026Methodology
Published / tracked
22 / 323
Verified
1
Provisional
21
Categories with evidence
7 / 8

Evidence by category

  • Agentic4 benchmarks
    Reported
  • Coding4 benchmarks
    Reported
  • Reasoning2 benchmarks
    Reported
  • Knowledge8 benchmarks
    Reported
  • Math1 benchmark
    Verified
  • Multilingual0 benchmarks
    Not measured
  • Multimodal1 benchmark
    Reported
  • Inst. Following2 benchmarks
    Reported
Open WeightSelf-hostNon-Reasoning
Confidence:
Low
snapshot

DeepSeek V3 ranks #147 out of 200 models on the public leaderboard with an overall score of 44.97/100. It also ranks #72 out of 99 on the verified leaderboard. While not a frontier model, it offers specific advantages depending on the use case.

DeepSeek V3 is a open weight model with a 128K token context window. It processes queries without explicit chain-of-thought reasoning, offering faster response times and lower token usage.

This profile currently has 22 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 Agentic (#97), while its weakest is Coding (#101). This performance profile makes it particularly useful for coding agents, browser research, and computer-use workflows.

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 44.645.13

  1. Z-1
    Z
    #14345.13
    Z-1 is #143 with a score of 45.13.
    Compare
  2. Nemotron-4 15B
    NVIDIA
    #14445.08
    Nemotron-4 15B is #144 with a score of 45.08.
    Compare
  3. Seed 1.6 Flash
    ByteDance
    #14545.08
    Seed 1.6 Flash is #145 with a score of 45.08.
    Compare
  4. Mistral 8x7B
    Mistral
    #14644.99
    Mistral 8x7B is #146 with a score of 44.99.
    Compare
  5. DeepSeek V3Current model
    DeepSeek
    #14744.97
    DeepSeek V3 is #147 with a score of 44.97.
  6. Moonshot v1
    Moonshot AI
    #14844.79
    Moonshot v1 is #148 with a score of 44.79.
    Compare
  7. Seed-2.0-Mini
    ByteDance
    #14944.6
    Seed-2.0-Mini is #149 with a score of 44.6.
    Compare

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. Agentic19%
    Eligible cohort rank #97 of 119Category score 38.5
  2. Coding17%
    Eligible cohort rank #101 of 122Category score 42.6

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 #97 of 119Percentile 19thWeight 22%4 benchmarksReported38.5
CodingRank #101 of 122Percentile 17thWeight 20%4 benchmarksReported42.6
ReasoningRank Not rankedWeight 17%2 benchmarksReported48.7
KnowledgeRank Not rankedWeight 12%8 benchmarksReported61.8
MathRank Not rankedWeight 5%1 benchmarkVerified26.3
MultilingualWeight 7%0 benchmarksNot measuredNot measured
MultimodalWeight 12%1 benchmarkReportedScore pending
Inst. FollowingRank Not rankedWeight 5%2 benchmarksReported86.1

Self-host vs API cost

Estimates at 50,000 req/day · 1000 tokens/req average.

DeepSeek V3
API / mo$1,028
Self-host / mo$18,221
Break-even1.2B/day
Model the full break-even

Chatbot Arena performance

Scroll horizontally to inspect confidence intervals and vote counts.

Chatbot Arena Elo, confidence interval, and vote count by evaluation view
ViewEloConfidence intervalVotes
Text Overall1358±4.821,770
Coding1388±10.13,280
Math1311±10.82,721
Instruction Following1344±6.98,606
Creative Writing1349±10.03,623
Multi-turn1374±9.83,747
Hard Prompts1350±8.15,408
Hard Prompts (English)1362±9.73,637
Longer Query1375±10.63,123

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.

Agentic4 benchmarks
AA Agentic IndexReported

Artificial Analysis Agentic Index

1.6%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

22.8%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.
GDPval-AAReported

GDPval-AA normalized

0.0%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.
GDPval-AAReported
217Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
Coding4 benchmarks
LiveCodeBenchReported

LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code

37.6%Weighted 38%
Source: Reported upstream sourceProvenance: Reported row carried from an upstream public source. Displayable on BenchLM, but not treated as verified unless explicitly marked otherwise.
SWE-bench VerifiedReported

Software Engineering Benchmark Verified

42%Weighted 16%
Source: Reported upstream sourceProvenance: Reported row carried from an upstream public source. Displayable on BenchLM, but not treated as verified unless explicitly marked otherwise.
AA Coding IndexReported

Artificial Analysis Coding Index

23.0%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.
AA-SciCodeReported

Artificial Analysis SciCode

35.4%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.
Reasoning2 benchmarks
AA-LCRReported

Artificial Analysis Long Context Reasoning

29.0%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

0.0%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.
Knowledge8 benchmarks
MMLU-ProReported

Massive Multitask Language Understanding Professional

75.9%Weighted 30%
Source: Reported upstream sourceProvenance: Reported row carried from an upstream public source. Displayable on BenchLM, but not treated as verified unless explicitly marked otherwise.
GPQAReported

Graduate-Level Google-Proof Q&A

59.1%Weighted 7%
Source: Reported upstream sourceProvenance: Reported row carried from an upstream public source. Displayable on BenchLM, but not treated as verified unless explicitly marked otherwise.
Artificial Analysis Intelligence IndexReported
14.2%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-GPQA DiamondReported

Artificial Analysis GPQA Diamond

55.7%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-HLEReported

Artificial Analysis Humanity's Last Exam

3.6%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 IndexReported

Artificial Analysis Omniscience Index

-41.3%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-Omniscience AccuracyReported

Artificial Analysis Omniscience Accuracy

25.4%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-Omniscience Hallucination RateReported

Artificial Analysis Omniscience Hallucination Rate

89.4%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.
Math1 benchmark
FrontierMath v2 (Tiers 1-3)Benchmark exact

FrontierMath v2 Tiers 1-3

1.724%Weighted 30%
Source: Epoch AI FrontierMath v2 leaderboardProvenance: Epoch AI reports FrontierMath v2 Tiers 1-3 at 1.724% for DeepSeek-V3. BenchLM selects the highest published thinking effort for the model and stores the v2 benchmark slice separately.
Multimodal1 benchmark
Design Arena WebsiteReported

Design Arena Website Elo

1150Display 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. Following2 benchmarks
IFEvalReported

Instruction-Following Eval

86.1%Weighted 35%
Source: Reported upstream sourceProvenance: Reported row carried from an upstream public source. Displayable on BenchLM, but not treated as verified unless explicitly marked otherwise.
AA-IFBenchReported

Artificial Analysis IFBench

34.8%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.

Frequently Asked Questions

How does DeepSeek V3 perform overall in AI benchmarks?

DeepSeek V3 has 22 published benchmark scores on BenchLM, but it does not yet have enough non-generated coverage to receive a global overall rank.

Is DeepSeek V3 good for knowledge and understanding?

DeepSeek V3 has visible benchmark coverage in knowledge and understanding, but BenchLM does not currently assign it a global category rank there.

Is DeepSeek V3 good for coding and programming?

DeepSeek V3 ranks #101 out of 122 models in coding and programming benchmarks with an average score of 42.6. There are stronger options in this category.

Is DeepSeek V3 good for mathematics?

DeepSeek V3 has visible benchmark coverage in mathematics, but BenchLM does not currently assign it a global category rank there.

Is DeepSeek V3 good for reasoning and logic?

DeepSeek V3 has visible benchmark coverage in reasoning and logic, but BenchLM does not currently assign it a global category rank there.

Is DeepSeek V3 good for agentic tool use and computer tasks?

DeepSeek V3 ranks #97 out of 119 models in agentic tool use and computer tasks benchmarks with an average score of 38.5. There are stronger options in this category.

Is DeepSeek V3 good for multimodal and grounded tasks?

DeepSeek V3 has visible benchmark coverage in multimodal and grounded tasks, but BenchLM does not currently assign it a global category rank there.

Is DeepSeek V3 good for instruction following?

DeepSeek V3 has visible benchmark coverage in instruction following, but BenchLM does not currently assign it a global category rank there.

Is DeepSeek V3 open source?

Yes, DeepSeek V3 is an open weight model created by DeepSeek, meaning it can be downloaded and run locally or fine-tuned for specific use cases.

Does DeepSeek V3 have full benchmark coverage on BenchLM?

Not yet. DeepSeek V3 currently has 22 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 V3?

DeepSeek V3 has a published context window of 128K, which determines how much text it can process in a single interaction.

Last updated: July 21, 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.