evals.report
BenchmarksLabsCompareRun guidesIn the wild
DeepSeekDeepSeek

DeepSeek V3.2

DeepSeek · DeepSeek. Released Dec 1, 2025.

DeepSeek V3.2 is a model from DeepSeek in the DeepSeek family, released Dec 1, 2025. evals.report tracks 31 reported DeepSeek V3.2 benchmark scores across FrontierMath, Berkeley Function Calling Leaderboard, ARC-AGI-1, SWE-bench Pro, AIME (OTIS Mock), GPQA Diamond, SimpleQA Verified, ARC-AGI-2, and 23 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

Open31 results

Benchmark results 31

Compare this model
BenchmarkCategoryScoreMetricStatusDate
FrontierMathReasoning22.1%accuracyOfficialDec 1, 2025Details
Berkeley Function Calling LeaderboardTool use56.73%accuracyOfficialDec 1, 2025Details
ARC-AGI-1Reasoning57%accuracyOfficialDec 1, 2025Details
SWE-bench ProCoding15.56%% resolvedOfficialDec 1, 2025Details
AIME (OTIS Mock)Reasoning87.8%accuracyOfficialDec 1, 2025Details
GPQA DiamondReasoning83.4%accuracyOfficialDec 1, 2025Details
SimpleQA VerifiedOther27.5%accuracyOfficialDec 1, 2025Details
ARC-AGI-2Reasoning4.03%accuracyOfficialDec 1, 2025Details
LMArenaChat preference1423source-defined ratingOfficialDec 1, 2025Details
WeirdMLCoding39.5%average accuracyOfficialDec 1, 2025Details
Artificial Analysis Intelligence IndexReasoning32IndexOfficialDec 1, 2025Details
Epoch Capabilities IndexReasoning146.5IndexOfficialDec 1, 2025Details
Aider PolyglotCoding74.2%% correctOfficialDec 1, 2025Details
SWE-rebenchCoding60.9%Resolved rate (pass@1)UnverifiedDec 1, 2025Details
MMLU-ProReasoning86.2%accuracyVerifiedDec 1, 2025Details
GAIA: A Benchmark for General AI AssistantsAgents34.8%accuracyUnverifiedDec 1, 2025Details
BrowseCompAgents40.1%accuracyUnverifiedDec 1, 2025Details
AIME 2026Reasoning94.17%accuracyOfficialDec 1, 2025Details
GDPvalAgents1197EloOfficialDec 1, 2025Details
LiveCodeBenchCoding59.3%Pass@1UnverifiedDec 1, 2025Details
SciCodeCoding38.7%accuracyUnverifiedDec 1, 2025Details
Global-MMLUReasoning86.5%accuracyUnverifiedDec 1, 2025Details
WebDev ArenaChat preference1332EloVerifiedDec 1, 2025Details
EQ-Bench Creative Writing v3Chat preference1515EloVerifiedDec 1, 2025Details
Design ArenaChat preference1220EloVerifiedDec 1, 2025Details
MCP-UniverseTool use19.91%Overall Success RateVerifiedDec 1, 2025Details
SWE-bench MultilingualCoding59.0%% resolvedOfficialDec 1, 2025Details
MathArena HMMT February 2026Reasoning84.09%accuracyOfficialDec 1, 2025Details
FrontierMath Tier 4Reasoning2.1%accuracyOfficialDec 1, 2025Details
Vectara Hallucination LeaderboardOther6.3%Hallucination RateOfficialDec 1, 2025Details
Terminal-Bench 2.0Agents39.6%task successOfficialDec 1, 2025Details