What is DeepSeek V3.2-Exp?

DeepSeek V3.2-Exp is a large language model from DeepSeek, released on September 29, 2025. It accepts text input.

When was DeepSeek V3.2-Exp released?

DeepSeek released DeepSeek V3.2-Exp on September 29, 2025.

What is DeepSeek V3.2-Exp's context window?

DeepSeek V3.2-Exp has a 128K-token context window.

How does DeepSeek V3.2-Exp score on benchmarks?

DeepSeek V3.2-Exp has 17 cited benchmark scores, including Vending-Bench 2 (Andon Labs) 1034, WeirdML (Håvard Tveit Ihle) 39.5% and LMArena Elo 1424.9 Elo. Each score links to its source.

DeepSeek V3.2-Exp

DeepSeek·Operational·Released 2025-09-29text

Context128k

Max output-

Input / 1M-

Output / 1M-

Cached in-

Benchmarks17

Benchmarks

Cited public sources · rank is exact-benchmark, apples-to-apples

Vending-Bench 2 (Andon Labs)

#46/591034 ↗

WeirdML (Håvard Tveit Ihle)

#69/9239.5% ↗

LMArena Elo

#82/1411424.9 ↗

LMArena WebDev (LMArena)

#82/941271.8 ↗

Aider-Polyglot

#2/874.5% ↗

AIME 2025

#10/2489.3% ↗

BrowseComp

#20/2340.1% ↗

BrowseComp-zh

#2/347.9% ↗

GPQA-Diamond

#5/879.9% ↗

HMMT 2025

#1/283.6% ↗

Humanity's Last Exam (no tools)

#24/3419.8% ↗

LiveCodeBench

#4/1174.1% ↗

MMLU-Pro

#10/4985% ↗

SimpleQA (agentic tool use)

#1/297.1% ↗

SWE Verified

#2/267.8% ↗

SWE-bench Multilingual

#11/1557.9% ↗

Terminal-bench

#2/337.7% ↗

Compare

vs Gemini 4 Argon vs GPT-6.1 Sol vs Claude Sonnet 5.5 vs Muse Spark 1.3 vs GPT-6 Sol vs GPT-6 Luna vs Claude Opus 5.5 vs GPT-6 Astra

More from DeepSeek

DeepSeek V4.1 Flash DeepSeek V4 Pro DeepSeek V3.2 DeepSeek R1 DeepSeek V3 DeepSeek V4 Flash

DeepSeek V3.2-Exp FAQ

What is DeepSeek V3.2-Exp?: DeepSeek V3.2-Exp is a large language model from DeepSeek, released on September 29, 2025. It accepts text input.
When was DeepSeek V3.2-Exp released?: DeepSeek released DeepSeek V3.2-Exp on September 29, 2025.
What is DeepSeek V3.2-Exp's context window?: DeepSeek V3.2-Exp has a 128K-token context window.
How does DeepSeek V3.2-Exp score on benchmarks?: DeepSeek V3.2-Exp has 17 cited benchmark scores, including Vending-Bench 2 (Andon Labs) 1034, WeirdML (Håvard Tveit Ihle) 39.5% and LMArena Elo 1424.9 Elo. Each score links to its source.

Access DeepSeek V3.2-Exp and every other model through one endpoint with automatic failover: Respan gateway.