DataLearner logo
DE

DeepSeek V3.2-Exp

Reasoning modelDeepSeek VDeepSeek V3.2

DeepSeek-V3.2-Exp

Release date: 2025-09-29Updated: 2026-07-17 23:18:41.7572,309
Live demoGitHubHugging FaceCompare
Parameters
671B
Context length
128K
Chinese support
Not supported
Reasoning ability

DeepSeek V3.2-Exp is a reasoning model from DeepSeek-AI, released on 2025-09-29. It accepts text input and returns text output. The cataloged parameter count is 671B, with 37B active parameters per inference. The recorded context window is 128K, and the recorded maximum output is 64K. Cataloged capabilities include Reasoning model. The checkpoint is listed under the MIT license, with the license link included in the model references. Use the linked references to confirm current access, licensing, and provider-specific limits.

Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology

DeepSeek V3.2-Exp

Model basics

Reasoning traces
Supported
Thinking modes
Thinking Mode (Default)
Context length
128K tokens
Max output length
64K tokens
Model type
Reasoning model
Modality (in / out)
Text → Text
Release date
2025-09-29
Model file size
1342GB
MoE architecture
Yes
Total params / Active params
671B / 37B
Knowledge cutoff
No data
DeepSeek V3.2-Exp

Open source & experience

DeepSeek V3.2-Exp

Official resources

Paper
Paper
DataLearnerAI blog
No blog post yet
DeepSeek V3.2-Exp

API details

API speed
3/5
No public API pricing yet.
DeepSeek V3.2-Exp

Benchmark Results

DeepSeek V3.2-Exp currently shows benchmark results led by SimpleQA (1 / 47, score 97.10), Aider-Polyglot (11 / 59, score 74.20), MMLU Pro (26 / 132, score 85). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.

Thinking

General Knowledge

9 evaluations
Benchmark / mode
Score
Rank/total
MMLU Pro
85
26 / 132
MMLU Pro
84
38 / 132
GPQA Diamond
79.90
84 / 188
GPQA Diamond
74
103 / 188
LiveBench
Standard Mode
49.85
91 / 115
LiveBench
Thinking Mode
58.90
73 / 115
HLE
20.30
118 / 173
HLE
19.80
120 / 173
HLE
8.60
153 / 173

Common Sense

1 evaluations
Benchmark / mode
Score
Rank/total
SimpleQA
97.10
1 / 47

Coding and Software Engineer

3 evaluations
Benchmark / mode
Score
Rank/total
LiveCodeBench
74.10
42 / 123
LiveCodeBench
55
85 / 123
SWE-bench Verified
67.80
72 / 113

Math and Reasoning

2 evaluations
Benchmark / mode
Score
Rank/total
AIME2025
89.30
39 / 107
AIME2025
58
84 / 107

AI Agent - Tool Usage

2 evaluations
Benchmark / mode
Score
Rank/total
Terminal-Bench
37.70
14 / 35
Terminal-Bench
23
30 / 35

Agent Level Benchmark

5 evaluations
Benchmark / mode
Score
Rank/total
Aider-Polyglot
Standard Mode
70.20
17 / 59
Aider-Polyglot
Thinking Mode
74.20
11 / 59
τ²-Bench
66.70
27 / 43
τ²-Bench - Telecom
34
34 / 35
τ²-Bench - Telecom
34
34 / 35

Instruction Following

1 evaluations
Benchmark / mode
Score
Rank/total
IF Bench
54.10
28 / 31

AI Agent - Information Search

1 evaluations
Benchmark / mode
Score
Rank/total
BrowseComp
40.10
48 / 53
View benchmark analysisCompare with other models

Compare with other models

No curated comparisons for this model yet.

Want a custom combination? Open the compare tool

DeepSeek V3.2-Exp

Publisher

DeepSeek-V3.2-Exp

Model Overview

DeepSeek V3.2-Exp is a reasoning model from DeepSeek-AI, released on 2025-09-29.

It accepts text input and produces text output. Its cataloged capabilities include Reasoning model. The cataloged parameter count is 671B, with 37B active parameters per inference. The recorded context window is 128K, and the recorded maximum output is 64K.

The checkpoint is listed under the MIT license, with the license link included in the model references. The evaluation section contains 24 cataloged benchmark results with their recorded modes and scores. The page links 4 release, model-card, repository, or provider references for checking the underlying claims. Specifications, availability, and prices can change; undisclosed values are intentionally left unstated.

DeepSeek V3.2-Exp

FAQ

What is DeepSeek V3.2-Exp?

DeepSeek V3.2-Exp is a reasoning model from DeepSeek-AI, released on 2025-09-29. It accepts text input and returns text output. The cataloged parameter count is 671B, with 37B active parameters per inference. The recorded context window is 128K, and the recorded maximum output is 64K. Cataloged capabilities include Reasoning model. The checkpoint is listed under the MIT license, with the license link included in the model references. Use the linked references to confirm current access, licensing, and provider-specific limits.

What input and output modalities does DeepSeek V3.2-Exp support?

The current model record lists text as input and text as output.

What are the main recorded specifications for DeepSeek V3.2-Exp?

The cataloged parameter count is 671B, with 37B active parameters per inference. The recorded context window is 128K, and the recorded maximum output is 64K. Fields without a source-backed value remain undisclosed.

Why is no API price shown for DeepSeek V3.2-Exp?

Retired experimental snapshot; no current first-party hosted price is published for the exact snapshot.

Are benchmark results available for DeepSeek V3.2-Exp?

The evaluation section contains 24 cataloged benchmark results with their recorded modes and scores. Compare only results that use the same benchmark version and evaluation mode.

Is DeepSeek V3.2-Exp open source?

The checkpoint is listed under the MIT license, with the license link included in the model references. Review the linked license text before commercial or derivative use.

DataLearner on WeChat

Follow DataLearner on WeChat for AI model updates and research notes.

DataLearner WeChat QR code