DeepSeek V3.2: API price, context window and how to connect

DeepSeek V3.2 by DeepSeek: $0.28 per million input tokens, $0.42 per million output tokens, a 164K-token window. What it is good for, what to compare it with and how to connect it in SKILLOK.

Updated Copy as Markdown
In this article

In short#

A previous-generation DeepSeek model with sparse attention (DeepSeek Sparse Attention): economical in compute and still strong at reasoning and tool calling. A 164-thousand-token window.

Specifications#

Provider DeepSeek
Released 2025-12-01
Context window 164K tokens
Accepts text
Answers with text
Reasoning mode yes
Tool calling yes
Open weights yes
Identifier deepseek/deepseek-v3.2

Price#

Per 1M tokens In dollars
Input (question, context) $0.28
Output (answer) $0.42

For reference: a conversation of 20 questions of ~500 tokens and answers of ~800 tokens costs about $0.01. Models in reasoning mode spend more tokens on an answer than the text shows: the reasoning is paid for too.

What it is good for#

  • A proven model for simple agents
  • Step-by-step reasoning
  • Tasks without long context

Similar models#

Catalog models with the same kind of output, closest in the price of a typical request.

Model Input Output Window Price
Mistral Small 4 $0.15 $0.6 262K about the same
Llama 4 Maverick $0.188 $0.652 1.05M 1.1× more expensive
Qwen3.8 Flash $0.15 $0.47 1M 1.2× cheaper
Qwen3.8 Omni Flash $0.15 $0.47 1M 1.2× cheaper
GPT-6 Luna $0.1 $0.5 1.05M 1.3× cheaper

Other DeepSeek models#

How to connect in SKILLOK#

DeepSeek V3.2 works through OpenRouter: add an OpenRouter key in Settings → AI and pick the deepseek/deepseek-v3.2 model. It then becomes available in Space AI, in tasks and to the repository agent. In detail — Connecting a model in SKILLOK.

Where the data comes from#

Prices, the context window and capabilities come from the public OpenRouter catalog as of 2026-10-19. A provider may change the price — the exact figure is always on the model's OpenRouter page.

Was this helpful?