DeepSeek

DeepSeek V3.2

deepseek/deepseek-v3.2

JSON

Specification

Published by the vendor; the platform bills at the target's declared price.

DeepSeek V3.2

reasoning open weights tools

DeepSeek deepseek/deepseek-v3.2

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...

Input $/1M: $0.27
Output $/1M: $0.40
Context: 164K
Max output: 66K
Released: Dec 1, 2025
Modalities: text

Intelligence index-

Coding index44

Agentic index-

Cached input reads $0.13 per 1M · weights on Hugging Face as deepseek-ai/DeepSeek-V3.2

On this deployment

No routing target resolves to this model yet. Declare one and the router picks it up without retraining.

Not routable here yet

Add a target with metadata.model set to this id.

targets.yamlyaml

# targets.yaml - declare a target that runs DeepSeek V3.2
targets:
  - id: deepseek-v3-2
    kind: llm
    name: DeepSeek V3.2
    metadata:
      model: deepseek/deepseek-v3.2          # links the target to this catalogue entry
    capabilities:
      domains: [general]
      supports_tools: true
      context_window: 163840
    cost:
      usd_per_1k_tokens: 0.000334

Price among DeepSeek models

Input list price per million tokens; DeepSeek V3.2 is the 6th cheapest of 13 listed.

More from DeepSeek

Other models by the same vendor in the catalogue.