UnifyBench

RankingEffort benchmarks

DeepSeek-V3.2 · Reasoning effort benchmarks

Compare measured Low, Medium, High and other settings. Every number keeps its source.

DeepSeek-V3.2

29 measured results · 2 reported effort labels · 16 benchmark/harness combinations. Model evidence and sources →

Capability score evidence

DeepSeek-V3.2 · Reasoning: 17.0 overall comparison score. Meets the overall evidence requirements.

Which sources carry the weight?
  • Artificial Analysis · 100.0% direct comparison weight · independent

These shares describe the comparisons entering the fit, not fractions of the final score.

How sensitive is this score?

Remove one evaluator or family and refit the graph against the same overall panel. Values use the matching dataset audited 2026-09-12. These are sensitivity checks, not confidence intervals; ranking eligibility may also change.

Removed evidenceRecomputed scoreChange
Artificial AnalysisComparison path lost—
ARC Prize19.6+2.6
Datacurve14.8-2.2
LiveCodeBench16.9-0.1
Scale18.4+1.4
Terminal-Bench17.0+0.0
Cognition17.6+0.6
LiveBench17.2+0.2
SWE-rebench17.4+0.3
Epoch AI17.1+0.1
Vals AI16.5-0.5
Provider reports · Anthropic17.0+0.0
Provider reports · Moonshot AI17.0-0.0
Provider reports · Meta17.0-0.0
Provider reports · Mistral17.0+0.0
All provider reports17.0-0.0
Every benchmark family
Removed evidenceRecomputed scoreChange
arc-agi19.5+2.5
deepswe14.9-2.2
gdpval18.1+1.1
mmmu-pro17.9+0.9
omniscience16.2-0.8
epoch-game-puzzles16.4-0.6
frontiercode17.6+0.6
hle17.6+0.6
simpleqa17.5+0.5
aa-lcr16.6-0.4
swe-rebench17.4+0.3
apex-agents17.3+0.3
tau-bench16.8-0.3
mmlu-pro17.3+0.2
critpt16.8-0.2
ifbench17.2+0.2
livecodebench16.8-0.2
multi-swe-bench16.8-0.2
terminal-bench17.2+0.2
harvey16.8-0.2
livebench-instructions17.2+0.2
scicode17.2+0.2
gpqa16.9-0.2
livebench-language17.2+0.2
frontiermath17.2+0.1
swe-atlas-qna16.9-0.1
aime16.9-0.1
gmmlu17.1+0.1
gdp-pdf17.1+0.1
arc-agi-316.9-0.1
livebench-coding16.9-0.1
vibe-code16.9-0.1
livebench-reasoning16.9-0.1
vals-finance-agent17.0-0.1
swe-atlas-test-writing17.0-0.1
vals-legal-research17.0-0.1
automationbench17.0-0.1
briefcase17.0-0.0
vals-code-migration17.0-0.0
itbench17.0-0.0
enterprise-ops17.0-0.0
math50017.1+0.0
analyst-agent17.0+0.0
swe-atlas-refactoring17.0-0.0
livebench-math17.0+0.0
livebench-data17.0+0.0
vals-excel-modeling17.0-0.0
enigma-eval17.0+0.0
swe-bench-pro17.0+0.0
What evidence is missing from the fit?

Across all collected settings for this model: 32 observations, 27 contributing, 0 matched without graph weight and 5 excluded. Counts do not establish rank eligibility.

  • 2 · Benchmark family or source protocol has not been reviewed
  • 3 · No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately
Inspect every observation and exclusion →

Capability profile

One configuration, seven capabilities

Comparison score · 0–100
DeepSeek-V3.2 · Reasoning capability radarComparison scores from 0 to 100. Unknown capabilities have no point; lines stop at gaps. Hollow points are preliminary. Exact values and support labels follow the chart.AgenticHard reasoningCodingHuman prefKnowledgeMultimodalLong context50100DeepSeek-V3.2 · Reasoning · Agentic: 17.6 · SupportedDeepSeek-V3.2 · Reasoning · Hard reasoning: 21.5 · SupportedDeepSeek-V3.2 · Reasoning · Coding: 18.2 · SupportedDeepSeek-V3.2 · Reasoning · Knowledge: 21.2 · SupportedDeepSeek-V3.2 · Reasoning · Long context: 29.9 · Preliminary

Gaps are unknown, not zero. Hollow points are preliminary.

  • Agentic

    A17.6Supported

  • Hard reasoning

    A21.5Supported

  • Coding

    A18.2Supported

  • Human pref

    Unknown

  • Knowledge

    A21.2Supported

  • Multimodal

    Unknown

  • Long context

    A29.9Preliminary

Each profile belongs to the named configuration. Capability scores use a shared panel for all axes, distinct from the overall ranking panel; they are not accuracy percentages or direct head-to-head wins.

Compare this configuration →Explore the model evidence profile →

These are published benchmark scores. Supported effort estimates also appear on the Capability leaderboard. Results from different harnesses occupy different rows. Missing cells stay unknown; no scores are copied between effort levels.

Benchmark / sourceReasoningNon-reasoning
AIME 2025Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

92%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

59%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

APEX-Agents-AAArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

14.53%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

Not reported
AA-LCR v1.1Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

73.33%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

45.67%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

CritPtArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Benchmark developed by Argonne and UIUC, with contributions from 60+ researchers globally · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

2.86%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

0.86%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

GDPval-AA v2 LeaderboardArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Elo rating for performance on real-world work tasks · Anchored to a human baseline of 1,000 · Higher is better · Evaluation results measured independently by Artificial Analysis

Unit: elo. Source collected 2026-09-12.

810.24 Elo
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

Not reported
Global-MMLU-LiteArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

86.54%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

82.68%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

GPQA DiamondArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

84.04%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

75.05%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

Humanity's Last ExamArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

24.56%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

11.21%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

IFBenchArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

60.68%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

49.05%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

LiveCodeBenchArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

86.24%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

59.26%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

MMLU-ProArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

86.21%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

83.71%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

AA-Omniscience IndexArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

AA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct. · Evaluation results measured independently by Artificial Analysis

Unit: index. Source collected 2026-09-12.

-22.48
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

-46.88
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

AA-Omniscience AccuracyArtificial Analysis · 6000 questions; 1 attempt; correct-answer share; GPT-5.6 Luna medium grader
Benchmark details

AA-Omniscience Accuracy (higher is better) measures the proportion of correctly answered questions out of all questions, regardless of whether the model chooses to answer · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

32.97%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

24%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

𝜏²-Bench TelecomArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

90.64%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

78.95%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

Terminal-Bench HardArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

35.61%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

32.58%
Reported settings

DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2

Terminal-Bench v2.1Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Benchmark developed by the Laude Institute and open-source contributors · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

46.82%
Reported settings

DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning

Not reported
What these comparisons mean

Each source reports its own effort labels. Matching labels do not establish equal compute budgets or identical fallback behavior. Open Reported settings to inspect the exact model name, agent and task coverage. Multiple results at one level remain visible rather than selecting the best. Source-reported None or Non-reasoning labels are not treated as evidence that an API supports those settings.

These measurements use their own dated source collection. The leaderboard fits identified configurations jointly; it contains no pooled model ratings. Explicit thinking budgets are separate configurations. Read the methodology.

Download all effort benchmark results · 7,424 results across 240 models · Collected 2026-09-12