Model

DeepSeek V4 Pro

DeepSeek · open weights

deepseek.com

Knowledge & Reasoning

Knowledge & Reasoning
BenchmarkScoreMetricHarnessSourceConfidenceLink
AI Consilium2.5mean_judge_scorestructuredIndependentmediumarxiv.org
CL-bench Life13.5overallN/AIndependentlowepoch.ai
CritPt12.9accuracyN/AIndependentlowepoch.ai
GPQA Diamond90.9accuracyN/AIndependenthighepoch.ai
HouseholdBench1.0relmaeN/AIndependentmediumarxiv.org
LexPilot0.6lca_macro-f1N/AIndependentmediumarxiv.org
MOF-VERIFY62.8macro-f1N/AIndependentmediumarxiv.org
PerSo62.0accuracyN/AIndependentmediumarxiv.org
SimpleQA Verified47.0accuracyN/AIndependenthighepoch.ai

Math

Math
BenchmarkScoreMetricHarnessSourceConfidenceLink
Australian SOC Benchmark76.1t2_accuracyN/AIndependentmediumarxiv.org

Coding

Coding
BenchmarkScoreMetricHarnessSourceConfidenceLink
A2Z GameSpec-Bench50.4gdd_fidelityN/AIndependentmediumarxiv.org
AAArena2288.2elocodexIndependentmediumarxiv.org
ArchitectureIQ68.4accuracyarchitectureiqIndependentmediumarxiv.org
ArrivalBench45.3failure_rateN/AIndependentmediumarxiv.org
MetaKernelBench25.7paired_liftcute dslIndependentmediumarxiv.org
ModeBench0.3pcmdN/AIndependentmediumarxiv.org
SCCGBench3.8mockN/AIndependentmediumarxiv.org
SciCode50.0scoreN/AIndependentmediumartificialanalysis.ai
SimuVerity28.4scoreclaude codeIndependentmediumarxiv.org
SpecGuardBench66.7accuracyopencodeIndependentmediumarxiv.org
SWE-bench Verified77.6resolved_rateN/AIndependenthighepoch.ai
SWE-Duel1.5bradley_terry_log_strengthmini-swe-agentIndependentmediumarxiv.org

Agentic / Tool Use

Agentic / Tool Use
BenchmarkScoreMetricHarnessSourceConfidenceLink
AgentDecisionBench7.2success_rateN/AIndependentmediumarxiv.org
AgenticBBO-Bench0.5hpoN/AIndependentmediumarxiv.org
AgentMonBench0.5silentswapN/AIndependentmediumarxiv.org
APTInvestBench50.8coverage_rateN/AIndependentmediumarxiv.org
CSEG-Bench0.8exactN/AIndependentmediumarxiv.org
DataSense-Bench36.5accuracyN/AIndependentmediumarxiv.org
DyadMem66.1accuracyN/AIndependentmediumarxiv.org
EC2-Bench72.9completionN/AIndependentmediumarxiv.org
HelpMeOut0.7m1N/AIndependentmediumarxiv.org
HICO-Bench59.2success_rateN/AIndependentmediumarxiv.org
IGMBench31.8success_rateN/AIndependentmediumarxiv.org
Legal Research Bench23.2accuracyN/AIndependentmediumarxiv.org
LifeSearch46.9scoreN/AIndependentmediumarxiv.org
OfficeTown Bench18.5pass@1N/AIndependentmediumarxiv.org
PraxisMemBench49.8success_rateN/AIndependentmediumarxiv.org
StoreBench0.7compositeN/AIndependentmediumarxiv.org
Synthetic Trading Benchmark8.2e0_sharpeN/AIndependentmediumarxiv.org
XiangqiBench5.0pass@3N/AIndependentmediumarxiv.org

Multimodal

Multimodal
BenchmarkScoreMetricHarnessSourceConfidenceLink
ChemArgus25.0full_problemtext onlyIndependentmediumarxiv.org
R-GroundBench71.0accuracyN/AIndependentmediumarxiv.org

Instruction Following

Instruction Following
BenchmarkScoreMetricHarnessSourceConfidenceLink
Nonlinear Narrative Generation Benchmark0.5acc.N/AIndependentmediumarxiv.org
UXBench Pro0.0turn_btN/AIndependentmediumarxiv.org
WISE55.0complianceN/AIndependentmediumarxiv.org

Safety & Alignment

Safety & Alignment
BenchmarkScoreMetricHarnessSourceConfidenceLink
AgentPrivArena19.3leak_rate_(%,_↓)agentprivarenaIndependentmediumarxiv.org
DecepEval80.6deception_rateN/AIndependentmediumarxiv.org
ObligationBench33.6recallN/AIndependentmediumarxiv.org
OverAct2.4sirN/AIndependentmediumarxiv.org
ProtoArena35.0success_rateopenhandsIndependentmediumarxiv.org
SWE-DeceptionBench74.2active_deception_rateN/AIndependentmediumarxiv.org
Trapped in Your Story99.8reversal_any_(%)N/AIndependentmediumarxiv.org

Multilingual

Multilingual
BenchmarkScoreMetricHarnessSourceConfidenceLink
CSWBench16.6cswbench_extended_/20N/AIndependentmediumarxiv.org
EHR-Bench86.0constructed_baN/AIndependentmediumarxiv.org
HakemBench0.6compositeN/AIndependentmediumarxiv.org

Reproduced by the community

Loading community runs…