DeepSeek: DeepSeek V3.2 Exp

DeepSeek-V3.2-Exp is an experimental large language model released by DeepSeek as an intermediate step between V3.1 and future architectures. It introduces DeepSeek Sparse Attention (DSA), a fine-gra…

deepseek/deepseek-v3.2-exp

DeepSeek-V3.2-Exp is an experimental large language model released by DeepSeek as an intermediate step between V3.1 and future architectures. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism designed to improve training and inference efficiency in long-context scenarios while maintaining output quality. Users can control the reasoning behaviour with the `reasoning` `enabled` boolean. [Learn more in our docs](https://openrouter.ai/docs/use-cases/reasoning-tokens#enable-reasoning-with-default-config)

The model was trained under conditions aligned with V3.1-Terminus to enable direct comparison. Benchmarking shows performance roughly on par with V3.1 across reasoning, coding, and agentic tool-use tasks, with minor tradeoffs and gains depending on the domain. This release focuses on validating architectural optimizations for extended context lengths rather than advancing raw task accuracy, making it primarily a research-oriented model for exploring efficient transformer designs.

Specifications & capabilities

Input
text
Output
text
Context window · tokens
163840
Knowledge cutoff
2025-07-31
Published model ID
DeepSeek-V3.2-Exp
Capabilities
reasoning, structuredOutputs, toolCalling
Description source ↗

Evaluations

41–60 / 102

LMArena

1,419.328

source id
lmarena
leaderboard name
LMArena Text Polish
dimension
general
metric
Arena rating
rank
105
sample size
439
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1391.7941384082098
confidence upper
1446.862561017726
Original result ↗

BFCL

34.85%

source id
bfcl
leaderboard name
BFCL Non-Live AST
dimension
toolUse
metric
Non-live AST accuracy
rank
105
Original result ↗

LMArena

1,410.528

source id
lmarena
leaderboard name
LMArena Text Style Control Industry Writing And Literature And Language
dimension
style
metric
Arena rating
rank
107
sample size
2655
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1399.0813375502214
confidence upper
1421.9753713116118
Original result ↗

LMArena

1,379.6

source id
lmarena
leaderboard name
LMArena Text Style Control Korean
dimension
style
metric
Arena rating
rank
107
sample size
211
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1339.714271987823
confidence upper
1419.4865408231526
Original result ↗

LMArena

1,461.887

source id
lmarena
leaderboard name
LMArena Text Chinese
dimension
general
metric
Arena rating
rank
108
sample size
1031
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1443.9162030272153
confidence upper
1479.8577819437257
Original result ↗

LMArena

1,419.677

source id
lmarena
leaderboard name
LMArena Text Factuality Non English
dimension
factuality
metric
Arena rating
rank
108
sample size
6570
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1412.0236603595
confidence upper
1427.3305907467484
Original result ↗

LMArena

1,456.845

source id
lmarena
leaderboard name
LMArena Text Factuality Hard Prompts
dimension
factuality
metric
Arena rating
rank
109
sample size
6247
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1449.1220266613668
confidence upper
1464.567050444467
Original result ↗

LMArena

1,444.389

source id
lmarena
leaderboard name
LMArena Text Factuality Longer Query
dimension
factuality
metric
Arena rating
rank
109
sample size
2082
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1432.3694729498366
confidence upper
1456.4088562426884
Original result ↗

LMArena

1,414.089

source id
lmarena
leaderboard name
LMArena Text Exclude Ties
dimension
general
metric
Arena rating
rank
110
sample size
8396
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1405.3308330004857
confidence upper
1422.8470043254094
Original result ↗

LMArena

1,433.607

source id
lmarena
leaderboard name
LMArena Text French
dimension
general
metric
Arena rating
rank
111
sample size
227
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1394.1763126375458
confidence upper
1473.0383788988638
Original result ↗

LMArena

1,372.26

source id
lmarena
leaderboard name
LMArena Text Korean
dimension
general
metric
Arena rating
rank
111
sample size
211
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1332.6558786657108
confidence upper
1411.8643650244787
Original result ↗

LMArena

1,441.471

source id
lmarena
leaderboard name
LMArena Text Style Control Longer Query
dimension
style
metric
Arena rating
rank
112
sample size
3052
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1430.7856779723722
confidence upper
1452.156030587741
Original result ↗

LMArena

1,422.402

source id
lmarena
leaderboard name
LMArena Text Overall
dimension
general
metric
Arena rating
rank
112
sample size
11955
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1416.0576103899443
confidence upper
1428.7458536129293
Original result ↗

LMArena

1,448.394

source id
lmarena
leaderboard name
LMArena Text Factuality Industry Life And Physical And Social Science
dimension
factuality
metric
Arena rating
rank
113
sample size
1185
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1432.4175539378618
confidence upper
1464.36978530989
Original result ↗

LMArena

1,437.528

source id
lmarena
leaderboard name
LMArena Text Hard Prompts English
dimension
general
metric
Arena rating
rank
114
sample size
2753
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1426.4690476516878
confidence upper
1448.5873758191988
Original result ↗

LMArena

1,424.044

source id
lmarena
leaderboard name
LMArena Text Factuality Instruction Following
dimension
factuality
metric
Arena rating
rank
114
sample size
2308
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1412.3006777305077
confidence upper
1435.7868215840033
Original result ↗

LMArena

1,429.389

source id
lmarena
leaderboard name
LMArena Text Hard Prompts
dimension
general
metric
Arena rating
rank
115
sample size
6557
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1421.483129709064
confidence upper
1437.2940315391659
Original result ↗

LMArena

1,418.086

source id
lmarena
leaderboard name
LMArena Text Longer Query
dimension
general
metric
Arena rating
rank
115
sample size
3052
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1407.4211105735803
confidence upper
1428.7514000362269
Original result ↗

LMArena

1,469.646

source id
lmarena
leaderboard name
LMArena Text Factuality Industry Software And It Services
dimension
factuality
metric
Arena rating
rank
116
sample size
2968
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1458.6958034340455
confidence upper
1480.5966416821464
Original result ↗

LMArena

1,463.374

source id
lmarena
leaderboard name
LMArena Text Factuality Hard Prompts English
dimension
factuality
metric
Arena rating
rank
116
sample size
1842
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1450.4440929097443
confidence upper
1476.304038051832
Original result ↗

Published prices

1–6 / 6

deepseek-ai/DeepSeek-V3.2-Exp ↗

$0.27

Inference provider
siliconFlow
Platform
SiliconFlow
Unit
input_tokens_per_million
Currency
USD
Billing conditions
Billing label
Input tokens (per 1M)

deepseek-ai/DeepSeek-V3.2-Exp ↗

$0.41

Inference provider
siliconFlow
Platform
SiliconFlow
Unit
output_tokens_per_million
Currency
USD
Billing conditions
Billing label
Output tokens (per 1M)

Provider specifications & access

1–3 / 3

DeepSeek-V3.2-Exp ↗

source model id
deepseek-ai/DeepSeek-V3.2-Exp
hosting
siliconflow-hosted
platforms
  • SiliconFlow
max output tokens
163840
architecture
input_modalities
  • text
output_modalities
  • text
capabilities
  • reasoning
  • structuredOutputs
  • toolCalling

deepseek-ai/DeepSeek-V3.2-Exp ↗

source model id
publishers/deepseek-ai/models/deepseek-v3-2@deepseek-v3-2-exp
hosting
vertex-self-deployed
platforms
  • Google Cloud Vertex AI
architecture
input_modalities
  • text
output_modalities
  • text
capabilities

    deepseek-ai/DeepSeek-V3.2-Exp ↗

    source model id
    publishers/hf-deepseek-ai/models/deepseek-v3.2-exp@001
    hosting
    vertex-self-deployed
    platforms
    • Google Cloud Vertex AI
    architecture
    input_modalities
    • text
    output_modalities
    • text
    capabilities

      Safety assessments

      1–1 / 1

      Foundation Model Transparency Index 2025 ↗

      32 / 100

      source id
      stanfordFMTI
      canonical model id
      deepseek/deepseek-v3.2-exp
      taxonomy
      transparency
      hazard category
      transparency
      hazard name
      Organization transparency
      policy area
      DeepSeek
      assessment type
      organization-level disclosure
      metric
      disclosure_score
      result label
      32 / 100
      sample size
      100
      updated at
      2026-11-09T18:08:05.227Z

      Source specification observations

      1–5 / 5

      openRouter ↗

      field
      context_length
      value
      163840
      normalized value
      163840
      scope
      source-model
      source id
      openRouter
      confidence
      1

      siliconFlow ↗

      field
      context_length
      value
      164K
      scope
      source-model
      source id
      siliconFlow
      confidence
      1

      openRouter ↗

      field
      knowledge_cutoff
      reported field
      knowledge_cutoff
      value
      2025-07-31
      normalized value
      2025-07-31
      precision
      day
      scope
      source-model
      source id
      openRouter
      confidence
      1

      openRouter ↗

      field
      max_output_tokens
      value
      147456
      normalized value
      147456
      scope
      source-model
      source id
      openRouter
      confidence
      1

      siliconFlow ↗

      field
      max_output_tokens
      value
      164K
      scope
      source-model
      source id
      siliconFlow
      confidence
      1

      Original sources

      1–19 / 19

      DeepSeek · Model reference ↗

      SiliconFlow · Model page ↗

      Hugging Face · Model repository ↗

      DeepInfra · Model page ↗

      Google Cloud · Model page ↗

      Google Cloud · Model page ↗

      OpenRouter · Model page ↗

      SiliconFlow · Playground ↗

      SiliconFlow · API documentation ↗

      GitHub · Model reference ↗

      DeepInfra · Catalog ↗

      Epoch AI · Catalog ↗

      GitHub · Catalog ↗

      GitHub · Catalog ↗

      Google Cloud · Catalog ↗

      LMArena · Catalog ↗

      OpenRouter · Catalog ↗

      SiliconFlow · Catalog ↗

      gorilla.cs.berkeley.edu · Catalog ↗

      Source links

      1–11 / 11

      Explore more models

      Models

      Loading
      EVIDENCE / COMPARISON

      Benchmarks

      Explore every benchmark

      Loading

      These charts contain model-linked results. The charts above also include unlinked source profiles. Compare scores within each evaluation. Model quality, usage, transparency and system performance measure different things.

      Install TensorTrace