supermarket

Language

DeepSeek-V4-Flash

By DeepSeek, in the DeepSeek-V family.

Released
24 Apr 2026
Status
Legacy
Context
1M
Max output
—
Input per 1M tokens
—
Output per 1M tokens
—
Open weights
Yes
License
mit
Parameters
284B, 13B active
Knowledge cutoff
—
Reasoning
Yes
Modalities
text → text

Category scores

Out of 100, from independent and third-party results only, by method category-2026-09-29.1.

Category scores
CategoryScoreRankCoverage
Knowledge & reasoning44.0293 of 4
Coding & software engineering18.8Not ranked1 of 5
Factuality & hallucination33.6481 of 1
Human preference64.6Not ranked2 of 5

Benchmark results

Every result we hold for this model, with who ran it and where it was published.

Knowledge & reasoning

Knowledge & reasoning results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
Chess PuzzlesPuzzles solved33.0%Independentmax2 Aug 2026Epoch AI
DTBench (Decision Theory Benchmark)Accuracy (capability questions)90.9%Third partymax—https://conceptualreasoning.ai/dtbench via Epoch AI
DTBench (Decision Theory Benchmark)Accuracy (capability questions)86.4%Third partymax—https://conceptualreasoning.ai/dtbench via Epoch AI
LMCA (Language Model Conceptual Argumentation)Correlation with expert ratings (Pearson r)0.4173Third partymax—https://conceptualreasoning.ai/lmca via Epoch AI
LMCA (Language Model Conceptual Argumentation)Correlation with expert ratings (Pearson r)0.3588Third partymax—https://conceptualreasoning.ai/lmca via Epoch AI
Mystery Game PuzzlesPuzzles solved34.0%Independentmax5 Aug 2026Epoch AI
SimpleBenchScore (AVG@5)61.1%Third partyunknown—https://lmcouncil.ai/benchmarks via Epoch AI
SimpleBenchScore (AVG@5)46.3%Third partyunknown—https://lmcouncil.ai/benchmarks via Epoch AI

Scientific reasoning

Scientific reasoning results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
GPQA DiamondAccuracy91.0%Independentmax2 Aug 2026Epoch AI

Mathematics

Mathematics results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
FrontierMath Tier 4Accuracy24.4%Independentmax2 Aug 2026Epoch AI
FrontierMath Tiers 1-3Accuracy57.5%Independentmax2 Aug 2026Epoch AI
OTIS Mock AIME 2024-2025Accuracy94.4%Independentmax2 Aug 2026Epoch AI

Coding & software engineering

Coding & software engineering results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
FrontierCodeMain score (Mean@5)18.8%Third partyunknown—https://cognition.com/frontiercode via Epoch AI
WeirdML (v2)Average accuracy63.0%Third partymax—https://htihle.github.io/weirdml.html via Epoch AI
WeirdML (v2)Average accuracy57.0%Third partyhigh—https://htihle.github.io/weirdml.html via Epoch AI
WeirdML (v2)Average accuracy45.6%Third partymax—https://htihle.github.io/weirdml.html via Epoch AI
WeirdML (v2)Average accuracy43.8%Third partyhigh—https://htihle.github.io/weirdml.html via Epoch AI

Factuality & hallucination

Factuality & hallucination results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
SimpleQA VerifiedAccuracy (% correct)33.6%Independentmax27 Aug 2026Epoch AI

Human preference

Human preference results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
Code Arena: WebDev (arena.ai)Arena score (Elo scale)1581Independenthigh25 Sep 2026Arena
Code Arena: WebDev (arena.ai)Votes5067Independenthigh25 Sep 2026Arena
Text Arena (arena.ai)Arena score (Elo scale)1436Independent—25 Sep 2026Arena
Text Arena (arena.ai)Votes51871Independent—25 Sep 2026Arena

Other DeepSeek-V versions

Sources