Skip to content
VECTOR WIREAI INTELLIGENCE
⌘K
UTC
Models741Deals30d268Policy30d67, 15 pendingSources1,014

DeepSeek-V3.2-Speciale

DeepSeekReleased Dec 1, 2025Legacy

Score basis

DeepSeek-V3.2-Speciale is behind the leaders in long context, reasoning, coding, instruction following, and factuality. Too few results yet to rate agentic tasks, 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.

Limited5 capabilities
  1. Long Context−25.3%1 of 3
  2. Reasoning−25.9%4 of 6
  3. Coding−29.5%3 of 10
  4. Instruction Following−32.3%1 of 3
  5. Factuality−34.4%2 of 4
Not rated5 capabilities

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

Price

$0.29input$0.43outputper million tokens

From deepseek · All prices

Cheaper than 70% of 331 priced models · 3:1 input-to-output blend, log scale

Evidence

27results on25benchmarks

  • 5 independently verified
  • 15 aggregator
  • 7 vendor-reported

From 7 sources · latest Oct 8, 2026 · How verification works

API features

Structured outputsReasoning

As listed by OpenRouter

DeepSeek-V3.2-Speciale benchmark results

27 results on 25 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

LimitedFull long context ranking

25.3% behind the leader1 of 3 ranked benchmarks measured

Reasoning

LimitedFull reasoning ranking

25.9% behind the leader4 of 6 ranked benchmarks measured

Show 1 more reasoning resultHide 1 reasoning result

Coding

LimitedFull coding ranking

29.5% behind the leader3 of 10 ranked benchmarks measured

Instruction Following

LimitedFull instruction following ranking

32.3% behind the leader1 of 3 ranked benchmarks measured

Factuality

LimitedFull factuality ranking

34.4% behind the leader2 of 4 ranked benchmarks measured

Agentic

Not enough dataFull agentic ranking

0 of 7 ranked benchmarks measured

More results

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

Show 8 more resultsHide 8 results

DeepSeek-V3.2-Speciale: common questions

Who makes DeepSeek-V3.2-Speciale?

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

When was DeepSeek-V3.2-Speciale released?

DeepSeek-V3.2-Speciale was released on Dec 1, 2025, according to Artificial Analysis.

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

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

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

DeepSeek-V3.2-Speciale costs $0.29 per million input tokens and $0.43 per million output tokens, according to deepseek. At a mix of three input tokens to one output token, it is cheaper than 70% of the 331 priced models we track.

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

We track 27 results for DeepSeek-V3.2-Speciale on 25 benchmarks from 7 sources, 5 of them independently verified. The latest was recorded on Oct 8, 2026.

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

OpenRouter lists structured outputs and reasoning for DeepSeek-V3.2-Speciale.

About this record

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

Tracked since
Apr 27, 2026
Newest source mention
Aug 23, 2026

Where the results come from

Verification: 27 scores · 5 independently verified · 15 aggregator-attributed · 7 vendor-reported. How these tiers are assigned

From 7 sources on 4 sites. Artificial Analysis supplies 15 of them; the 5 independently verified results come from 2 sites. Bars are coloured by trust tier.

  • artificialanalysis.ai15
  • api.llm-stats.com7
  • matharena.ai4
  • simple-bench.com1

Also known as

deepseek 3.2 speciale