DeepSeek-V3.2

DeepSeek-V3.2 places in the top three on 7 benchmarks.

DeepSeek Released December 1st, 2025 Family DeepSeek-V3 The lab's model card Compare with others

170

Benchmarks with a score

Ranked first, out of five or more models

In the top three

DeepSeek Dec 2025

DeepSeek-V3.2

Open weights Flagship

146.3 #90 of 256

Epoch Capabilities Index

Self-Reported Index 146.5 #121 of 404

SWE-Bench Pro 15.6% #20 /24
Terminal-Bench 2.0 39.5% #24 /43
Toolathlon 35.2% #29 /50

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
MMLU-Pro	85%	#30 of 167	Lab model card
Vectara Hallucination Leaderboard (2025 dataset)	6.30%	#27 of 104	Official leaderboard
LMArena (Elo)	1,425	#98 of 335	Official leaderboard
KINA	38%	#12 of 37	Official leaderboard
Creative Writing v3	71.7	#45 of 120	Official leaderboard
HUMAINE	29	#19 of 50	Official leaderboard
Humanity's Last Exam	40.8%	#64 of 154	Lab model card
Longform Creative Writing (EQ-Bench)	66.8	#50 of 119	Official leaderboard
LOCA-Bench	6.70%	#4 of 9	Official leaderboard
Judgemark v2.1	66.3	#25 of 42	Official leaderboard
HealthBench-Hallu	6.10%	#2 of 3	Official leaderboard
SkateBench v2	46.7%	#15 of 21	Official leaderboard
Extended NYT Connections	36.7%	#79 of 103	Official leaderboard
BullshitBench	14.6%	#86 of 107	Official leaderboard
Antidote	954	#29 of 36	Official leaderboard
CL-bench Life	9.50%	#13 of 16	Official leaderboard
HLE-Verified (text-only Full)	36.4%	#5 of 6	Official leaderboard
BullshitBench V2	13%	#108 of 129	Official leaderboard
HLE Extended	8.10%	#27 of 31	Official leaderboard
Role-play Benchmark (MiniMax)	60.3	#7 of 8	Official leaderboard
Lech Mazur Creative Writing	-2.92	#45 of 50	Official leaderboard
CL-bench	13.2%	#22 of 24	Official leaderboard
DesignArena ASCII Art	1,077	#71 of 77	Official leaderboard
AA-LCR Extended	24.7%	#48 of 49	Official leaderboard

Math and science

Benchmark	Score	Rank	Source
HiPhO	65.7%	#5 of 36	Official leaderboard
ChemBench	65.3%	#5 of 24	Official leaderboard
BRUMO 2025	96.7%	#9 of 42	Official leaderboard
AIME 2025	93.1%	#32 of 143	Lab model card
HMMT 2025	92.5%	#19 of 54	Lab model card
LiveMathematicianBench	29%	#6 of 17	Official leaderboard
OTIS Mock AIME 2024-2025	87.8%	#76 of 212	Epoch AI run
GPQA Diamond	83.4%	#86 of 229	Epoch AI run
DTBench	85.6%	#69 of 180	Official leaderboard
FrontierMath	22.1%	#27 of 70	Epoch AI run
SMT 2025	87.7%	#16 of 40	Official leaderboard
General365	54.9%	#11 of 23	Official leaderboard
CMIMC 2025	84.4%	#17 of 33	Official leaderboard
HMMT Nov 2025	90%	#12 of 22	Official leaderboard
A^3-Bench	56%	#4 of 7	Official leaderboard
FrontierMath Tier 4	2.10%	#37 of 55	Epoch AI run
ARC-AGI	57%	#64 of 95	Official leaderboard
AIME 2026	94.2%	#21 of 31	Official leaderboard
MathArena (Apex, IMO, USAMO)	2.08%	#32 of 47	Official leaderboard
ARC-AGI-2	4.03%	#74 of 98	Official leaderboard
IMO-AnswerBench	78.3%	#29 of 37	Lab model card
MathArena Apex Shortlist	50.5%	#30 of 38	Official leaderboard
ProofBench	8.00%	#47 of 59	Official leaderboard
HMMT 2026	84.1%	#25 of 31	Official leaderboard
Leipzig Benchmark	8.00%	#17 of 19	Official leaderboard
Echo General AI Prediction Leaderboard	964	#16 of 17	Official leaderboard

Coding

Benchmark	Score	Rank	Source
LiveCodeBench	83.3%	#7 of 54	Lab model card
SWE-bench Verified	70%	#15 of 60	Official leaderboard
Codeforces rating	2,386	#12 of 47	Lab model card
LiveCodeBench v6	83.3%	#24 of 90	Lab model card
ABC-Bench	50.1%	#2 of 7	Official leaderboard
AInsteinBench	37%	#2 of 6	Official leaderboard
CF-Div2-Stepfun	81.6%	#3 of 8	Official leaderboard
StructEval	75.5%	#14 of 28	Official leaderboard
OctoCodingbench	26%	#3 of 6	Official leaderboard
PRDBench	40.1%	#2 of 4	Official leaderboard
IOI (Vals)	14.4%	#30 of 55	Official leaderboard
DesignArena Website	1,168	#91 of 151	Official leaderboard
SWE-bench Multilingual	70.2%	#44 of 73	Lab model card
DesignArena 3D Design	1,115	#86 of 136	Official leaderboard
OpenHands Index	45.7%	#21 of 33	Official leaderboard
DesignArena Code	1,154	#94 of 143	Official leaderboard
DesignArena UI Component	1,140	#93 of 139	Official leaderboard
DesignArena Game Dev	1,122	#98 of 143	Official leaderboard
FrontierCS	24.8	#10 of 14	Official leaderboard
OTelBench	11.6%	#10 of 14	Official leaderboard
Code Arena	1,362	#96 of 128	Official leaderboard
OJBench	51.7%	#6 of 8	Official leaderboard
DesignArena Data Visualization	1,156	#89 of 118	Official leaderboard
SWE-Review	6.40%	#4 of 5	Official leaderboard
SWE-Bench Pro	15.6%	#20 of 24	Official leaderboard
DesignArena SVG	1,023	#86 of 103	Official leaderboard
WebDev Arena	1,325	#89 of 105	arena.ai, via Epoch AI
FeatureBench	35.9%	#14 of 16	Official leaderboard
FeatureBench Lite	35.9%	#14 of 16	Official leaderboard
Vibe Code Bench	5.11%	#106 of 120	Official leaderboard
APEX-SWE	25.6%	#59 of 65	Official leaderboard
SWE Atlas Codebase QnA Extended	0%	#36 of 38	Official leaderboard
SWE-bench Verified Extended	10%	#42 of 44	Official leaderboard

Agents and tools

Benchmark	Score	Rank	Source
Agent Poker Bench	1,090	#4 of 17	Official leaderboard
OdysseyArena Repo System	48.3%	#2 of 8	Official leaderboard
WebArena	74.3%	#1 of 4	Official leaderboard
LisanBench	2348.3	#40 of 133	Official leaderboard
OdysseyArena Turn on Lights	18.3%	#3 of 8	Official leaderboard
VITA-Bench	24%	#7 of 18	Official leaderboard
tau2-bench	80.3%	#33 of 79	Lab model card
BalatroBench	8.67	#7 of 16	Official leaderboard
LLM Chess	189	#56 of 123	Official leaderboard
ClawProBench	60.1%	#23 of 49	Official leaderboard
OdysseyArena AI Trading	8.62%	#4 of 8	Official leaderboard
BrowseComp-ZH	65%	#10 of 19	Lab model card
Terminal-Bench 2.0	39.5%	#24 of 43	Official leaderboard
Toolathlon	35.2%	#29 of 50	Official leaderboard
BrowseComp	67.6%	#48 of 81	Lab model card
GAIA	37.2%	#38 of 64	Official leaderboard
DeepPlanning	27.4%	#9 of 15	Official leaderboard
AutoResearchBench	4.21%	#8 of 13	Official leaderboard
SnakeBench	20	#134 of 188	Official leaderboard
RWSearch	74%	#5 of 7	Official leaderboard
WorldTravel	14%	#5 of 7	Official leaderboard
Claw-Eval-Live	51.4%	#9 of 12	Official leaderboard
EnterpriseOps-Gym	24.2%	#9 of 12	Official leaderboard
Pencil Puzzle Bench	0%	#40 of 53	Official leaderboard
DeR2 Bench	75.3%	#4 of 5	Official leaderboard
PinchBench	59.9%	#39 of 48	Official leaderboard
Game Arena Poker	-16.86	#26 of 31	Official leaderboard
WildClawBench	34%	#26 of 31	Official leaderboard
AgencyBench	28.6%	#8 of 9	Official leaderboard
MCPMark	38%	#9 of 10	Lab model card
Terminal-Bench 2.1 Extended	15.2%	#40 of 44	Official leaderboard
QwenClawBench	50.7%	#11 of 12	Official leaderboard
Game Arena Five in a Row	126	#19 of 20	Official leaderboard
BrowseComp Extended	3.30%	#40 of 42	Official leaderboard
Game Arena Chess	86	#40 of 42	Official leaderboard
Claw-Eval	67.5%	#23 of 24	Official leaderboard
Claw-Eval Pass^3	40.2%	#23 of 24	Official leaderboard
Game Arena Word Association	15	#39 of 40	Official leaderboard
Game Arena Checkers	59	#40 of 41	Official leaderboard
Game Arena Four in a Row	9	#42 of 43	Official leaderboard
Kaggle Game Arena	0	#42 of 42	Official leaderboard
Game Arena Dark Hex	0	#41 of 41	Official leaderboard
Game Arena Dots and Boxes	0	#40 of 40	Official leaderboard
Clocktower Radio	1,213	#23 of 23	Official leaderboard
Game Arena Chess Openings	0	#11 of 11	Official leaderboard
OdysseyArena Energy Dispatch	0%	#8 of 8	Official leaderboard
EnvSimBench	32.5%	#7 of 7	Official leaderboard
MCP-Universe	45.9%	#1 of 1	Lab model card

Professional work

Benchmark	Score	Rank	Source
BizFinBench v2	41.7%	#3 of 9	Official leaderboard
OccuBench	69.6%	#5 of 14	Official leaderboard
ProfBench	50.5%	#28 of 72	Official leaderboard
SpreadsheetBench 2	15.6%	#5 of 8	Official leaderboard
PLawBench	58%	#7 of 11	Official leaderboard
CaseLaw v2	55.4%	#34 of 53	Official leaderboard
TaxEval v2	68.2%	#93 of 129	Official leaderboard
FoodTruck Bench	$2,058	#44 of 59	Official leaderboard
LegalBench	76.1%	#103 of 138	Official leaderboard
CorpFin v2	51%	#94 of 119	Official leaderboard
Vending-Bench 2	$1,034	#54 of 67	Official leaderboard
EnterpriseBench: CoreCraft	24.1%	#42 of 51	Official leaderboard
APEX-1	51.5%	#67 of 71	Official leaderboard
Vals Finance Agent	2.35%	#48 of 49	Official leaderboard
APEX-Agents	21.3%	#60 of 61	Collected by Epoch AI, source not named
APEX-Agents 1.1	35.1%	#64 of 65	Official leaderboard
Vals Index v1	26.3%	#34 of 34	Official leaderboard

Vision, audio and robotics

Benchmark	Score		Rank	Source
MineBench	1,037		#66 of 75	Official leaderboard

Safety and security

Benchmark	Score	Rank	Source
BioTIER Permit	99.9%	#4 of 69	Official leaderboard
nocot-bench (NCRI)	75.0	#76 of 185	Official leaderboard
LiveSecBench	56.2	#17 of 34	Official leaderboard
Lakera B3 (Backbone Breaker Benchmark)	51.9%	#23 of 35	Official leaderboard
EvalDetectBench	0.80	#7 of 10	Official leaderboard
KORA	24.3%	#31 of 42	Official leaderboard
World-Class Bio	32.1%	#59 of 74	Official leaderboard
BinaryAudit	12.3%	#16 of 20	Official leaderboard
Human Pathogen Capabilities Test	40%	#70 of 86	Official leaderboard
Human Pathogen Capabilities Test v2 (HPCT v2)	37.4%	#58 of 71	Official leaderboard
SMDD-Bench	24.3%	#6 of 7	Official leaderboard
Molecular Biology Capabilities Test	39%	#80 of 90	Official leaderboard
Agentic BAIM-LLM Evaluation (ABLE)	23.8%	#16 of 18	Official leaderboard
HelpBench	80%	#8 of 9	Official leaderboard
Dictatorship Eval	0.97%	#19 of 20	Official leaderboard
BioTIER Refuse	10.2%	#68 of 69	Official leaderboard

Other

Benchmark	Score	Rank	Source
ACCoRD	64.6	#75 of 152	Official leaderboard
Conceptual Reasoning Index	45.4	#87 of 151	Official leaderboard
LMCA	28.8%	#97 of 147	Official leaderboard
CAIS Text Capabilities Index	21.5	#50 of 68	Official leaderboard
CAIS Risk Index	68.3	#64 of 67	Official leaderboard

Its lab's numbers put it at 146.5, in line with its ECI

The Self-Reported Index fits one score to the 17 capability results DeepSeek reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 144.8 to 148.3. How we compute it.

Its lab's numbers next to 11 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark	Lab says	Board says	Gap
SWE-bench Multilingual	70.2% SWE Multilingual (Resolved)	59% Leaderboard, mini-SWE-agent	+11 pts
Terminal-Bench 2.0	46.4% Terminal Bench 2.0 (Acc)	39.5% Leaderboard, Terminus 2	+6.9 pts
SWE-bench Verified	73.1% SWE Verified (Resolved)	70% Leaderboard, high, mini-SWE-agent	+3.1 pts
MCPMark	38% MCP-Mark (Pass@1)	36.8% Leaderboard, thinking	+1.2 pts
AIME 2025	93.1% AIME 2025 (Pass@1)	94.2% Leaderboard, thinking	−1.1 pts
GPQA Diamond	82.4% GPQA Diamond (Pass@1)	83.4% Epoch AI run, Thinking	−1.0 pts
HMMT Nov 2025	90.2% HMMT Nov 2025 (Pass@1)	90% Leaderboard, thinking	+0.2 pts
tau2-bench	80.3% τ2-Bench (Pass@1)	80.4% Leaderboard	Same
HMMT 2025	92.5% HMMT Feb 2025 (Pass@1)	92.5% Leaderboard, thinking	Same
MMLU-Pro	85% MMLU-Pro (EM)	85% Leaderboard	Same
Toolathlon	35.2% Tool-Decathlon (Pass@1)	35.2% Leaderboard	Same

DeepSeek's models over time

DeepSeek's models in color, against every other model on the Epoch Capabilities Index, by release date.

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

DeepSeek-V3.2 technical report, Table 2 Dec 1, 2025

MMLU-Pro (EM) 85
GPQA Diamond (Pass@1) 82.4
Best of HLE (Pass@1) / HLE (Pass@1), Search Agent section 40.8
LiveCodeBench (Pass@1-COT) 83.3
Codeforces (Rating) 2386
AIME 2025 (Pass@1) 93.1
HMMT Feb 2025 (Pass@1) 92.5
HMMT Nov 2025 (Pass@1) 90.2
IMOAnswerBench (Pass@1) 78.3
Terminal Bench 2.0 (Acc) 46.4
SWE Verified (Resolved) 73.1
SWE Multilingual (Resolved) 70.2
BrowseComp (Pass@1) 67.6
BrowseCompZh (Pass@1) 65
τ2-Bench (Pass@1) 80.3
MCP-Universe (Success Rate) 45.9
MCP-Mark (Pass@1) 38
Tool-Decathlon (Pass@1) 35.2

Its system card

Every result in the longer documents the lab published with it. 1 of them isn't in the launch post.

DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models (arXiv 2512.02556) Dec 1, 2025