DeepSeek-V3.2

Language Proprietary

Benchmarks

About DeepSeek-V3.2

DeepSeek-V3.2 is an AI model developed by DeepSeek, in the language category, released in 2025, made available as a proprietary (API-only) model. DeepSeek V3.2 is the official successor to V3.2-Exp and DeepSeek's flagship reasoning-first model built for agents. It o.

On this page you'll find DeepSeek-V3.2's full specifications, including 22 benchmark results. Review provider pricing and benchmark scores below, or compare DeepSeek-V3.2 head-to-head with other language models.

→ More models by DeepSeek → All language models → AI benchmarks leaderboard

Links

Announcement ↗ Paper ↗

Community

Would you run DeepSeek-V3.2 again?

No votes yet — be the first.

Did it run? — community reports by hardware

No reports yet. Own this model on some hardware? Be the first to confirm it runs.

Discussion

No comments yet. Start the discussion.

DeepSeek-V3.2 benchmark scores▶

Compare on the benchmark table →

Benchmark	Category	Score	Variant
BrowseComp	agentic	51.4%	thinking
BrowseComp-Zh	agentic	65	thinking
MCP-Mark	agentic	38	thinking
MCP-Universe	agentic	45.9	thinking
Terminal-Bench 2.0	agentic	46.4%	cited-devstral-readme
Terminal-Bench 2.0	agentic	46.4%	thinking
Tool-Decathlon	agentic	35.2	thinking
τ²-Bench (TAU2)	agentic	80.3	thinking
Codeforces Elo	code	2,386	thinking-rating
LiveCodeBench	code	83.3%	thinking
SWE-bench Verified	code	73.1%	cited-devstral-readme
SWE-bench Verified	code	73.1%	thinking
SWE-Bench Multilingual	coding	70.2	thinking
SWE-Bench Multilingual	coding	70.2	cited-devstral-readme
MMLU-Pro	general	85%	thinking
AIME 2025	math	93.1%	thinking
HMMT February 2025	math	92.5%	thinking
HMMT November 2025	math	90.2	thinking
IMOAnswerBench	math	78.3	thinking
GPQA Diamond	reasoning	82.4%	thinking
Humanity's Last Exam	reasoning	40.8%	thinking-search-agent
Humanity's Last Exam	reasoning	25.1%	thinking-text-only

More models like DeepSeek-V3.2

Llama-3-70B-Synthia-v3.5

· 70B

Frequently asked questions

What is DeepSeek-V3.2? ▶

DeepSeek V3.2 is the official successor to V3.2-Exp and DeepSeek's flagship reasoning-first model built for agents. It o It is tracked on GenAIList with its specifications, benchmark scores and provider pricing.

Who created DeepSeek-V3.2? ▶

DeepSeek-V3.2 was developed by DeepSeek and released in 2025.

Is DeepSeek-V3.2 open source or proprietary? ▶

DeepSeek-V3.2 is a proprietary model. It is accessed through an API rather than by downloading the weights.

How much does DeepSeek-V3.2 cost? ▶

Pricing for DeepSeek-V3.2 depends on the provider. See the providers table on this page for the latest API rates.

How does DeepSeek-V3.2 perform on benchmarks? ▶

DeepSeek-V3.2 is benchmarked across 22 evaluations on GenAIList, including BrowseComp (51.4%). See the full benchmark table below and compare it with other models.