DeepSeek V3.2

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...

Compare

Context window: 164K
Input: Text
Output: Text
Released: 2025-12-01

Max output: 66K
Knowledge cutoff: 2024-07
Model parameters: 685B
Weights: Open
License: mit
Capabilities: Open weight, Reasoning, Structured, Temperature, Tool call

Pricing

No official pricing available.

Estimate usage cost

Compare verified public prices for this exact model and region. Enter uncached input and cache usage separately. Estimates exclude unspecified fees, credits and currency conversion. Set requests to zero to estimate token charges only.

Input tokens

Output tokens

Cache read tokens

Cache write tokens

Requests

Context tokens (optional)

Currency

Batch processing

Performance

Overall capability

43.0/100

Coding: —
Reasoning: 46.6
Professional work: 29.5
Knowledge & language: 45.1


Vibe Code Bench v1.1	Coding	Vals AI	2026-11-05
Humanity’s Last ExamDeepSeek V3.2 (Reasoning)	Reasoning	Artificial Analysis	2026-11-05
IFBench（Artificial Analysis）DeepSeek V3.2 (Reasoning)	Knowledge & language	Artificial Analysis	2026-11-05
APEX-Agents 1.1	Professional work	Mercor	2026-11-05
GDPval-AA	Professional work	Artificial Analysis	2026-11-05
AA-LCRDeepSeek V3.2 (Reasoning)	Knowledge & language	Artificial Analysis	2026-11-05
CritPtDeepSeek V3.2 (Reasoning)	Reasoning	Artificial Analysis	2026-11-05
BullshitBench v2	Knowledge & language	BullshitBench	2026-11-05

Providers

							Cache write
OpenRouter	deepseek/deepseek-v3.2	Openai chat	global				-

ModelRegistry · Catalog updated 2026-11-05