DeepSeek V3.2

ChatInput: TextReleased Dec 1, 2025

Built by DeepSeek · China · deepseek.com

DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...

How to use

model: "deepseek/deepseek-v3.2"

Cap quality here and save

Hard questions go to this model, easy ones drop to something cheaper. Set a savings target (0–0.8) and the router tunes itself until realized savings converge on it.

{
  "model": "celon/auto",
  "celon": {
    "anchor": "deepseek/deepseek-v3.2",
    "savings_target": 0.4
  }
}

Naming a model directly gets you that model — savings controls don't apply.

Context

164K

Max output

66K

Input $/1M

$0.27

Output $/1M

$0.40

Features

Tool callingJSON

Serving providers

DeepInfraNovitaFriendli

Providers

Price, latency, and uptime per provider serving this model. Celon routes each request to the fastest and cheapest of them. Click a row for regions and data policies.

Loading provider metrics…

Performance

Measured latency (lower is better) and throughput (higher is better) per provider.

Speed

Loading provider metrics…