Skip to content
DeepSeek

DeepSeek V3.2

deepseek/deepseek-v3.2
Open weightsLanguage

Frontier-adjacent open-weight flagship at roughly a tenth of typical flagship cost, with reliable tool calling.

Model overview

Context window

128K

tokens

Input price

—

quoted on request

Output price

—

quoted on request

Weekly volume

—

tokens / week

Open weights

Yes

self-hostable

Capabilities

5

of 11 tags

Description

DeepSeek V3.2 is a language model from DeepSeek. Webparam does not stock this model yet and can source it on request — tell us what you need it for and we will come back with availability and a price. DeepSeek’s stated focus is open-weight frontier reasoning at disruptive economics. Its mean it can also be self-hosted under its licence.

Capabilities

Tool calling
Calls functions you define and returns their arguments as structured data, the basis of agents.
JSON mode
Constrains the response to valid JSON matching a schema you supply.
Streaming
Emits tokens as they are generated, so answers appear progressively rather than all at once.
Caching
Reuses already-processed prompt prefixes across requests, cutting input cost for repeated context.
Open weights
The parameters are published, so the model can be inspected, fine-tuned and self-hosted under its licence.

Strengths & weaknesses

Strengths

  • Frontier-adjacent quality at ~1/10 typical flagship cost
  • Strong tool-calling reliability
  • Open weights permit self-hosting later

Weaknesses

  • English long-form prose trails Western flagships
  • No vision input
  • Reasoning line (R1) is a separate model

Pricing

Pricing for DeepSeek V3.2, per 1M tokens
RatePriceUnit
Input— quoted on request—
Output— quoted on request—

Prompt caching is supported, so repeated prefixes bill below the listed input rate. This model is not provisioned yet; the rate is quoted on request.

Context window

128Ktokens

covers prompt and response together, so a long input leaves less room for the answer.

128K tokens against a peer range of 32K to 10M, median 256K. Compared across 34 language models. Logarithmic scale.

Supported features

Feature support for DeepSeek V3.2
FeatureSupport
Tool callingSupported
JSON modeSupported
StreamingSupported
Vision inputNot supported
Audio inputNot supported
Long contextNot supported
Open weightsSupported
ReasoningNot supported
Fine-tunableNot supported
BatchNot supported
CachingSupported

Example use cases

  • Bulk summarisation

  • Agent loops

  • Cost-sensitive production

Example prompt

prompt
Extract every commitment, owner and deadline from this meeting transcript as JSON.

Related models

Comparison

Compare against anything →
DeepSeek V3.2 compared with DeepSeek R1 and Qwen3 235B A22B Instruct 2507
AttributeDeepSeek V3.2DeepSeek R1Qwen3 235B A22B Instruct 2507
Context window128K164K256K
Input priceR0.00R0.00R0.00
Output price— not charged— not charged— not charged
Tool callingSupportedSupportedSupported
JSON modeSupportedNot supportedSupported
Vision inputNot supportedNot supportedNot supported
Open weightsSupportedSupportedSupported
Prompt cachingSupportedNot supportedNot supported

Best value in each row is highlighted. Blank cells are facts the supplier does not publish, not zeroes.

Provider information

View DeepSeek →
DeepSeek5 models

DeepSeek · 深度求索 · Hangzhou, China · est. 2023

Spun out of a quantitative trading firm, DeepSeek built its reputation by releasing open-weight models that matched closed frontier quality at a fraction of the price. Its V-series redefined cost expectations for production LLMs, and its R-series brought open reasoning models into serious enterprise conversations.

Recent releases

    Documentation

    Frequently asked questions

    Input is billed at R0.00 per 1M tokens, and there is no separate output charge. Prompt caching bills repeated prefixes below the listed input rate. Every figure on this page is an illustrative placeholder.