DeepSeek V3

DeepSeek V3 is a non-reasoning model from DeepSeek in the DeepSeek family. 24 benchmarks count toward its score, in 6 categories.

availableShows if the model has enough results for an index.
IndexOverall score. 50 is the middle.32.6 ±4.8
CoverageShare of the index weight with results.85%
SpeedOutput tokens per second.
Input / 1MUS dollars per 1M input tokens.$0.27
Output / 1MUS dollars per 1M output tokens.$1.1
ContextMaximum tokens in one request.128K
EloLMArena rating and rank.1332 (#209)

50 is the middle of the board. The range shows the doubt in the index.

21,770 votes. Elo shows what people prefer. It does not change the score.

CapabilitiesScore per category. 50 is the middle.

50 is the middle
AgenticMulti-step tasks with tools.
29.5
CodingCode writing and repair.
37.2
ReasoningLogic problems and puzzles.
33.5
MultimodalTasks with images and text.
N/A
KnowledgeFacts and expert knowledge.
34.2
MultilingualTasks in many languages.
N/A
InstructionTasks with strict rules in the prompt.
30.6
MathMath problems.
31.4

Results

24 counted
BenchmarkThe test name.CategoryThe capability that the test measures.ResultThe score from the publisher.IndexThis result on the index scale.RunThe settings of the run.DateDate of the result.Published byThe source of the result.
MGSMMultilingual91.9%9 Jan 2026Vals AI
MGSMMultilingual91.7%9 Jan 2026Vals AI
MATH 500Math88.6%40.89 Jan 2026Vals AI
Instruction-Following EvalInstruction86.1%36.8Jeffrey Zhou et al.
MATH 500Math84.5%36.29 Jan 2026Vals AI
MMLU ProKnowledge79.5%45.71 Sept 2026Vals AI
MMLU ProKnowledge76.6%41.31 Sept 2026Vals AI
Massive Multitask Language Understanding ProfessionalKnowledge75.9%40.1Yubo Wang et al.
LiveCodeBenchCoding65.5%43.91 Sept 2026Vals AI
LiveCodeBenchCoding65.5%43.91 Sept 2026Vals AI
MATH level 5Math64.9%30.5Epoch AI
GPQA DiamondKnowledge61.6%35.11 Sept 2026Vals AI
Graduate-Level Google-Proof Q&AKnowledge59.1%32.8David Rein et al.
GPQA DiamondKnowledge58.1%31.91 Sept 2026Vals AI
GPQA diamondKnowledge56.5%30.4Epoch AI
Artificial Analysis GPQA DiamondKnowledge55.7%25.6Artificial Analysis
AIMEMath52.2%38.116 Apr 2026Vals AI
Software Engineering Benchmark VerifiedCoding42.0%31.2Carlos E. Jimenez et al.
SWE-bench VerifiedCoding42.0%31.2SWE-Exp1 Sept 2026SWE-bench team
AIMEMath39.9%32.316 Apr 2026Vals AI
LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for CodeCoding37.6%18.5Naman Jain et al.
SWE-bench LiteCoding36.7%34.5KGCompass11 Sept 2025SWE-bench team
Artificial Analysis SciCodeCoding35.8%42.6Artificial Analysis
Artificial Analysis IFBenchInstruction34.8%24.4Artificial Analysis
SWE-bench LiteCoding30.7%29.4Moatless Tools11 Sept 2025SWE-bench team
SWE-bench LiteCoding30.7%29.4Moatless Tools11 Sept 2025SWE-bench team
Artificial Analysis Long Context ReasoningReasoning29.3%28.5Artificial Analysis
Artificial Analysis Omniscience AccuracyKnowledge25.5%45.4Artificial Analysis
Artificial Analysis Coding IndexCoding23.0%35.2Artificial Analysis
τ²-Bench Tool-Agent-User EvaluationAgentic22.8%15.2Victor Barres et al.
OTIS Mock AIME 2024-2025Math15.8%20.5Epoch AI
Artificial Analysis Intelligence IndexKnowledge8.5%33.0Artificial Analysis
Artificial Analysis Humanity's Last ExamKnowledge2.9%27.7Artificial Analysis
FrontierMath-2025-02-28-PrivateMath1.7%32.2Epoch AI
IOI v1Coding1.7%40.99 Aug 2026Vals AI
IOI v1Coding1.7%40.99 Aug 2026Vals AI
Artificial Analysis Agentic IndexAgentic0.8%37.8Artificial Analysis
GDPval-AA normalizedAgentic0.0%35.6Artificial Analysis
Critical Physics TasksReasoning0.0%38.4Artificial Analysis

24 benchmarks count, from 37 of 39 results. A grey row does not count. Too few models took that benchmark.

Sources

BenchLM benchmark aggregationUsed with attribution; per-benchmark results credited to their original authorsVals AI, collected directlyNo licence stated. Read from the public leaderboard and credited to Vals AIEpoch AI, collected directlyCC BY — free to use and redistribute with attributionSWE-bench team, collected directlyNo licence stated. The repository publishes submission records for reproducibility and transparency and asks that SWE-bench be cited

Same level, lower price

Qwen3.8-Flash-Next66.4 · Freedots3-note Preview65.5 · FreeApodex 1.1 Mini59.1 · Free

More from DeepSeek

DeepSeek V4.1 Flash70.6DeepSeek V4 Pro 081368.8DeepSeek V4 Flash 073164.6DeepSeek V4 Pro 042362.4DeepSeek V4 Flash 042363.0