Skip to content
VECTOR WIREAI INTELLIGENCE
⌘K
UTC
Models741Deals30d271Policy30d66, 14 pendingSources1,015

DeepSeek-V3.2-Exp

DeepSeekReleased Sep 29, 2025Deprecated

Score basis

DeepSeek-V3.2-Exp is capable in long context and factuality; and behind the leaders in reasoning, instruction following, coding, and agentic tasks. Too few results yet to rate safety, math, multimodal tasks, or multilingual tasks.

Capability profile

Bars show the model's median result as a share of the leading model's, per capability. Select a row to see its results.

Capable2 capabilities
  1. Long Context−22.5%1 of 3
  2. Factuality−24.7%3 of 4
Limited4 capabilities
  1. Reasoning−36.0%3 of 6
  2. Instruction Following−37.0%1 of 3
  3. Coding−39.8%5 of 10
  4. Agentic−41.1%2 of 7
Not rated4 capabilities

Too few results yet to rate Safety, Math, Multimodal or Multilingual.

Price

$0.28input$0.42outputper million tokens

From Artificial Analysis · 2 providers tracked · All prices

Cheaper than 70% of 330 priced models · 3:1 input-to-output blend, log scale

Evidence

50results on33benchmarks

  • 9 independently verified
  • 28 aggregator
  • 13 vendor-reported

From 8 sources · latest Oct 8, 2026 · How verification works

API features

Tool callingStructured outputsReasoning

As listed by OpenRouter

DeepSeek-V3.2-Exp benchmark results

50 results on 33 benchmarks, grouped by capability. Each bar is the result as a share of the capability leader's; every result links to its source.

Long Context

CapableFull long context ranking

22.5% behind the leader1 of 3 ranked benchmarks measured

Show 1 more long context resultHide 1 long context result

Factuality

CapableFull factuality ranking

24.7% behind the leader3 of 4 ranked benchmarks measured

Show 2 more factuality resultsHide 2 factuality results

Reasoning

LimitedFull reasoning ranking

36.0% behind the leader3 of 6 ranked benchmarks measured

Show 4 more reasoning resultsHide 4 reasoning results

Instruction Following

LimitedFull instruction following ranking

37.0% behind the leader1 of 3 ranked benchmarks measured

Show 1 more instruction following resultHide 1 instruction following result

Coding

LimitedFull coding ranking

39.8% behind the leader5 of 10 ranked benchmarks measured

Show 2 more coding resultsHide 2 coding results

Agentic

LimitedFull agentic ranking

41.1% behind the leader2 of 7 ranked benchmarks measured

Show 1 more agentic resultHide 1 agentic result

More results

Benchmarks outside the capability baskets. They are not ranked against a leader.

Show 18 more resultsHide 18 results

DeepSeek-V3.2-Exp: common questions

Who makes DeepSeek-V3.2-Exp?

DeepSeek-V3.2-Exp is made by DeepSeek.

When was DeepSeek-V3.2-Exp released?

DeepSeek-V3.2-Exp was released on Sep 29, 2025, according to Artificial Analysis.

What is DeepSeek-V3.2-Exp good at?

DeepSeek-V3.2-Exp is capable in long context and factuality; and behind the leaders in reasoning, instruction following, coding, and agentic tasks. Too few results yet to rate safety, math, multimodal tasks, or multilingual tasks.

How much does DeepSeek-V3.2-Exp cost?

DeepSeek-V3.2-Exp costs $0.28 per million input tokens and $0.42 per million output tokens, according to Artificial Analysis. We track its price at 2 providers. At a mix of three input tokens to one output token, it is cheaper than 70% of the 330 priced models we track.

How many benchmarks has DeepSeek-V3.2-Exp been tested on?

We track 50 results for DeepSeek-V3.2-Exp on 33 benchmarks from 8 sources, 9 of them independently verified. The latest was recorded on Oct 8, 2026.

Which API features does DeepSeek-V3.2-Exp support?

OpenRouter lists tool calling, structured outputs, and reasoning for DeepSeek-V3.2-Exp.

About this record

Where DeepSeek-V3.2-Exp's numbers come from, and every name it appears under.

Tracked since
Apr 27, 2026
Newest source mention
Sep 10, 2026

Where the results come from

Verification: 50 scores · 9 independently verified · 28 aggregator-attributed · 13 vendor-reported. How these tiers are assigned

From 8 sources on 6 sites. Artificial Analysis supplies 28 of them; the 9 independently verified results come from 4 sites. Bars are coloured by trust tier.

  • artificialanalysis.ai28
  • api.llm-stats.com13
  • raw.githubusercontent.com4
  • matharena.ai3
  • aider.chat1
  • datasets-server.huggingface.co1

Also known as

deepseek v3.2 exp (non-reasoning)deepseek v3.2 exp (reasoning)deepseek-v3.2-exp (reasoner)deepseek-v3.2-exp (chat)deepseek-v3.2-exp (think)deepseek-v3.2-exp-thinking