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

21–40 / 102

LMArena

1,407.525

source id
lmarena
leaderboard name
LMArena Text Industry Entertainment And Sports And Media
dimension
general
metric
Arena rating
rank
81
sample size
2177
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1395.1125341738473
confidence upper
1419.9364715187485
Original result ↗

LMArena

1,440.049

source id
lmarena
leaderboard name
LMArena Text Spanish
dimension
general
metric
Arena rating
rank
82
sample size
352
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1409.0055933989852
confidence upper
1471.091705827421
Original result ↗

LMArena

1,455.883

source id
lmarena
leaderboard name
LMArena Text Factuality English
dimension
factuality
metric
Arena rating
rank
83
sample size
4309
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1446.9135948044652
confidence upper
1464.8516478493643
Original result ↗

LMArena

1,446.356

source id
lmarena
leaderboard name
LMArena Text Style Control German
dimension
style
metric
Arena rating
rank
86
sample size
226
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1409.8101828855147
confidence upper
1482.9019056128354
Original result ↗

BFCL

67%

source id
bfcl
leaderboard name
BFCL Irrelevance Detection
dimension
toolUse
metric
Irrelevance detection
rank
86
Original result ↗

LMArena

1,422.28

source id
lmarena
leaderboard name
LMArena Text Factuality Industry Writing And Literature And Language
dimension
factuality
metric
Arena rating
rank
90
sample size
1777
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1409.6186467677937
confidence upper
1434.9406245760122
Original result ↗

LMArena

1,411.872

source id
lmarena
leaderboard name
LMArena Text Factuality Creative Writing
dimension
factuality
metric
Arena rating
rank
91
sample size
1079
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1395.846332349282
confidence upper
1427.8980688758193
Original result ↗

LMArena

1,403.643

source id
lmarena
leaderboard name
LMArena Text Creative Writing
dimension
general
metric
Arena rating
rank
91
sample size
1659
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1389.1716661962334
confidence upper
1418.114635869352
Original result ↗

LMArena

1,409.695

source id
lmarena
leaderboard name
LMArena Text Style Control Industry Entertainment And Sports And Media
dimension
style
metric
Arena rating
rank
94
sample size
2177
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1397.204369753227
confidence upper
1422.1856364911093
Original result ↗

LMArena

1,408.986

source id
lmarena
leaderboard name
LMArena Text Style Control Creative Writing
dimension
style
metric
Arena rating
rank
95
sample size
1659
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1394.3953319994557
confidence upper
1423.5759684567868
Original result ↗

LMArena

1,440.54

source id
lmarena
leaderboard name
LMArena Text English
dimension
general
metric
Arena rating
rank
96
sample size
5020
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1431.8521192989513
confidence upper
1449.2274862278664
Original result ↗

LMArena

1,437.495

source id
lmarena
leaderboard name
LMArena Text Factuality Exclude Ties
dimension
factuality
metric
Arena rating
rank
97
sample size
8205
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1429.6101735615475
confidence upper
1445.3798211033368
Original result ↗

LMArena

1,436.083

source id
lmarena
leaderboard name
LMArena Text Style Control Spanish
dimension
style
metric
Arena rating
rank
97
sample size
352
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1404.65972702121
confidence upper
1467.5065674426373
Original result ↗

LMArena

1,439.952

source id
lmarena
leaderboard name
LMArena Text Factuality Industry Business And Management And Financial Operations
dimension
factuality
metric
Arena rating
rank
99
sample size
1466
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1424.5170104437946
confidence upper
1455.3860936549709
Original result ↗

LMArena

1,443.824

source id
lmarena
leaderboard name
LMArena Text Factuality Industry Legal And Government
dimension
factuality
metric
Arena rating
rank
101
sample size
467
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1415.7202466996505
confidence upper
1471.9282956116383
Original result ↗

LMArena

1,407.381

source id
lmarena
leaderboard name
LMArena Text Industry Writing And Literature And Language
dimension
general
metric
Arena rating
rank
102
sample size
2655
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1395.9850442967754
confidence upper
1418.7774216185444
Original result ↗

LMArena

1,438.476

source id
lmarena
leaderboard name
LMArena Text Factuality Multi Turn
dimension
factuality
metric
Arena rating
rank
104
sample size
1326
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1422.7260805311316
confidence upper
1454.2253907524282
Original result ↗

BFCL

37.5%

source id
bfcl
leaderboard name
BFCL Relevance Detection
dimension
toolUse
metric
Relevance detection
rank
104
Original result ↗

LMArena

1,436.289

source id
lmarena
leaderboard name
LMArena Text Factuality Overall
dimension
factuality
metric
Arena rating
rank
105
sample size
11862
evaluated at
2026-11-09T00:00:00.000Z
source model id
DeepSeek-V3.2-Exp
confidence lower
1430.3707570431893
confidence upper
1442.207642809769
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