UAB

DeepSeek V3.2 Exp

Live · updated continuously

MenuDeepSeek V3.2 Exp

Find

Find a model Comparison table Score overview

Browse

Models Benchmarks Sources Compare

Method

Data status Data & sources Scoring method Methodology Changelog

Model profile · DeepSeek

DeepSeek V3.2 Exp

Open weightsbudget · registry tag 2026 preview

Thin verified coverage

Reads as thin verified coverage across the resolved source data.

Visible coverage: 12.3%
Verified coverage: 12.3%
Spread: n/a
Last verified: Aug 19, 2026

textcodedocument7 aliases33 official source links

Open compare

UAB

Live · updated continuously

MenuLoading

Find

Find a model Comparison table Score overview

Browse

Models Benchmarks Sources Compare

Method

Data status Data & sources Scoring method Methodology Changelog

Model detail

Loading model evidence.

Preparing the model profile, benchmark coverage, and registry evidence.

Chat / text26 benchmarks75.6%

Intelligence Index

AA · Chat / text · Combined

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #152 · Source label: DeepSeek V3.2 Exp (Non-reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 22
Percentile: 67.7%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `intelligenceIndex`.

67.7% percentile inside its fair comparison set

22Raw benchmark value

AA-Omniscience accuracy

AA · Chat / text · Objective

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #122 · Source label: DeepSeek V3.2 Exp (Non-reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 22.8%
Percentile: 66.9%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `omniscienceAccuracy`.

66.9% percentile inside its fair comparison set

22.8%Raw benchmark value

AA-Omniscience non-hallucination

AA · Chat / text · Objective

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #287 · Source label: DeepSeek V3.2 Exp (Non-reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 7.9%
Percentile: 21.9%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `omniscienceNonHallucination`.

21.9% percentile inside its fair comparison set

7.9%Raw benchmark value

IFBench

AA · Chat / text · Objective

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #156 · Source label: DeepSeek V3.2 Exp (Non-reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 43.1%
Percentile: 55.1%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `ifbench`.

55.1% percentile inside its fair comparison set

43.1%Raw benchmark value

Blended price

AA · Chat / text · Speed / cost

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #91 · Source label: DeepSeek V3.2 Exp (Reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: $0.3 /1M tokens
Percentile: 70.1%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `price1mBlended0To3To1`.

70.1% percentile inside its fair comparison set

$0.3 /1M tokensRaw benchmark value

Input price

AA · Chat / text · Speed / cost

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #118 · Source label: DeepSeek V3.2 Exp (Reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: $0.3 /1M input tokens
Percentile: 61.1%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `price1mInputTokens`.

61.1% percentile inside its fair comparison set

$0.3 /1M input tokensRaw benchmark value

Output price

AA · Chat / text · Speed / cost

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #75 · Source label: DeepSeek V3.2 Exp (Reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: $0.4 /1M output tokens
Percentile: 75.4%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `price1mOutputTokens`.

75.4% percentile inside its fair comparison set

$0.4 /1M output tokensRaw benchmark value

Openness Index

AA · Chat / text · Combined

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #85 · Source label: DeepSeek V3.2 Exp (Reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 44
Percentile: 68.5%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `opennessBreakdown.opennessIndex`.

68.5% percentile inside its fair comparison set

44Raw benchmark value

Text Arena

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #67 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,423
Percentile: 79.7%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: overall. Source rank: #85. Votes: 11922. Organization: deepseek. License: MIT.

79.7% percentile inside its fair comparison set

1,423Raw benchmark valueCI 1,417 - 1,429

Text Arena · Creative Writing

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #45 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,410
Percentile: 86.4%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: creative_writing. Source rank: #58. Votes: 1628. Organization: deepseek. License: MIT.

86.4% percentile inside its fair comparison set

1,410Raw benchmark valueCI 1,396 - 1,425

Text Arena · English

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #63 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,438
Percentile: 80.9%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: english. Source rank: #79. Votes: 5676. Organization: deepseek. License: MIT.

80.9% percentile inside its fair comparison set

1,438Raw benchmark valueCI 1,429 - 1,446

Text Arena · Exclude Ties

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #63 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,416
Percentile: 80.9%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: exclude_ties. Source rank: #80. Votes: 8383. Organization: deepseek. License: MIT.

80.9% percentile inside its fair comparison set

1,416Raw benchmark valueCI 1,407 - 1,425

Text Arena · Hard Prompts

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #59 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,448
Percentile: 82.2%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: hard_prompts. Source rank: #76. Votes: 6464. Organization: deepseek. License: MIT.

82.2% percentile inside its fair comparison set

1,448Raw benchmark valueCI 1,440 - 1,456

Text Arena · Hard Prompts English

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #65 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,454
Percentile: 80.2%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: hard_prompts_english. Source rank: #81. Votes: 3206. Organization: deepseek. License: MIT.

80.2% percentile inside its fair comparison set

1,454Raw benchmark valueCI 1,443 - 1,464

Text Arena · Instruction Following

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #62 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,417
Percentile: 81.2%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: instruction_following. Source rank: #78. Votes: 3310. Organization: deepseek. License: MIT.

81.2% percentile inside its fair comparison set

1,417Raw benchmark valueCI 1,406 - 1,427

Text Arena · Longer Query

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #55 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,442
Percentile: 82.2%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: longer_query. Source rank: #69. Votes: 2933. Organization: deepseek. License: MIT.

82.2% percentile inside its fair comparison set

1,442Raw benchmark valueCI 1,431 - 1,453

Text Arena · Multi Turn

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #59 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,431
Percentile: 82%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: multi_turn. Source rank: #76. Votes: 1984. Organization: deepseek. License: MIT.

82% percentile inside its fair comparison set

1,431Raw benchmark valueCI 1,418 - 1,444

Text Arena · No Style Control

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #57 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,423
Percentile: 82.8%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: overall. Source rank: #69. Votes: 11922. Organization: deepseek. License: MIT.

82.8% percentile inside its fair comparison set

1,423Raw benchmark valueCI 1,416 - 1,429

Text Arena · Creative Writing · No Style Control

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #46 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,404
Percentile: 86.1%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: creative_writing. Source rank: #59. Votes: 1628. Organization: deepseek. License: MIT.

86.1% percentile inside its fair comparison set

1,404Raw benchmark valueCI 1,390 - 1,419

Text Arena · English · No Style Control

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #52 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,440
Percentile: 84.3%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: english. Source rank: #64. Votes: 5676. Organization: deepseek. License: MIT.

84.3% percentile inside its fair comparison set

1,440Raw benchmark valueCI 1,431 - 1,448

Text Arena · Exclude Ties · No Style Control

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #55 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,415
Percentile: 83.4%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: exclude_ties. Source rank: #67. Votes: 8383. Organization: deepseek. License: MIT.

83.4% percentile inside its fair comparison set

1,415Raw benchmark valueCI 1,406 - 1,423

Text Arena · Hard Prompts · No Style Control

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #59 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,430
Percentile: 82.2%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: hard_prompts. Source rank: #73. Votes: 6464. Organization: deepseek. License: MIT.

82.2% percentile inside its fair comparison set

1,430Raw benchmark valueCI 1,422 - 1,438

Text Arena · Hard Prompts English · No Style Control

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #60 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,438
Percentile: 81.8%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: hard_prompts_english. Source rank: #71. Votes: 3206. Organization: deepseek. License: MIT.

81.8% percentile inside its fair comparison set

1,438Raw benchmark valueCI 1,428 - 1,449

Text Arena · Instruction Following · No Style Control

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #67 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,403
Percentile: 79.7%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: instruction_following. Source rank: #81. Votes: 3310. Organization: deepseek. License: MIT.

79.7% percentile inside its fair comparison set

1,403Raw benchmark valueCI 1,392 - 1,413

Text Arena · Longer Query · No Style Control

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #63 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,419
Percentile: 79.6%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: longer_query. Source rank: #76. Votes: 2933. Organization: deepseek. License: MIT.

79.6% percentile inside its fair comparison set

1,419Raw benchmark valueCI 1,408 - 1,429

Text Arena · Multi Turn · No Style Control

AR · Chat / text · Human

It tests whether the model is actually useful in normal conversational turns, not just on narrow correctness tasks.

Rank #59 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,424
Percentile: 82%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: multi_turn. Source rank: #73. Votes: 1984. Organization: deepseek. License: MIT.

82% percentile inside its fair comparison set

1,424Raw benchmark valueCI 1,411 - 1,437

Coding7 benchmarks52.6%

Terminal-Bench Hard

AA · Coding · Objective

It tells you whether the model can generate, repair, and reason over code under evaluator pressure rather than marketing examples.

Rank #86 · Source label: DeepSeek V3.2 Exp (Non-reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 25%
Percentile: 74.2%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `terminalbenchHard`.

74.2% percentile inside its fair comparison set

25%Raw benchmark value

SciCode

AA · Coding · Objective

It tells you whether the model can generate, repair, and reason over code under evaluator pressure rather than marketing examples.

Rank #127 · Source label: DeepSeek V3.2 Exp (Reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 37.7%
Percentile: 71.5%
Last updated: recent
Eligibility

Reasoning / math / science5 benchmarks73.6%

Humanity's Last Exam

AA · Reasoning / math / science · Objective

It is one of the cleaner reads on deliberate reasoning strength rather than style or popularity.

Rank #166 · Source label: DeepSeek V3.2 Exp (Non-reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 9%
Percentile: 62.7%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `hle`.

62.7% percentile inside its fair comparison set

9%Raw benchmark value

GPQA

AA · Reasoning / math / science · Objective

It is one of the cleaner reads on deliberate reasoning strength rather than style or popularity.

Rank #156 · Source label: DeepSeek V3.2 Exp (Non-reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 73.8%
Percentile: 65.6%
Last updated: recent
Eligibility: preview_model

Professional reasoning18 benchmarks77%

Text Arena · Expert

AR · Professional reasoning · Human

Observed user preference in Arena's Text Arena expert leaderboard.

Rank #87 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,422
Percentile: 68.7%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: expert. Source rank: #107. Votes: 609. Organization: deepseek. License: MIT.

68.7% percentile inside its fair comparison set

1,422Raw benchmark valueCI 1,399 - 1,446

Text Arena · Industry Business And Management And Financial Operations

AR · Professional reasoning · Human

Observed user preference in Arena's Text Arena industry_business_and_management_and_financial_operations leaderboard.

Rank #62 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,422
Percentile

Search / tool use1 benchmark48.8%

Tau2-Bench Telecom

AA · Search / tool use · Objective

It matters when the model must browse, call tools, and recover useful answers from external systems.

Rank #173 · Source label: DeepSeek V3.2 Exp (Reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 33.9%
Percentile: 48.8%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `tau2`.

48.8% percentile inside its fair comparison set

33.9%Raw benchmark value

Long context1 benchmark59.4%

Long Context Reasoning

AA · Long context · Objective

It checks whether long-context claims survive contact with retrieval, memory, or long-document tasks.

Rank #158 · Source label: DeepSeek V3.2 Exp (Non-reasoning)

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Artificial Analysis
Raw value: 45%
Percentile: 59.4%
Last updated: recent
Eligibility: preview_model

Parsed from Artificial Analysis public leaderboard field `lcr`.

59.4% percentile inside its fair comparison set

45%Raw benchmark value

Multilingual14 benchmarks75.6%

Text Arena · Chinese

AR · Multilingual · Human

Observed user preference in Arena's Text Arena chinese leaderboard.

Rank #54 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,475
Percentile: 82%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: chinese. Source rank: #66. Votes: 624. Organization: deepseek. License: MIT.

82% percentile inside its fair comparison set

1,475Raw benchmark valueCI 1,453 - 1,498

Text Arena · French

AR · Multilingual · Human

Observed user preference in Arena's Text Arena french leaderboard.

Rank #56 · Source label: deepseek-v3.2-exp

verified runtimeexact aliasBackground only

Raw row drilldownsource row, percentile, last updated, eligibility

Source: Arena
Raw value: 1,450
Percentile: 74.5%
Last updated: aging
Eligibility: preview_model

Parsed from Arena leaderboard dataset row `deepseek-v3.2-exp`. Category: french. Source rank: #71. Votes: 198. Organization: deepseek. License: MIT.

DeepSeek V3.2 Exp

Loading model evidence.

DeepSeek V3.2 Exp

Source-linked scores by benchmark

Source links and registry checks

Current snapshot.