Skip to content

DeepSeek V3.2

ModelOpen sourceActive
by DeepSeek · family “deepseek-v”

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... (description from the OpenRouter listing)

in: textout: textReasoningTool useStructured output
Epoch Capabilities Index
146.3 #70 of 268
90% CI 144.0 – 147.4
Listing ↗
Released
1 Dec 2025
Context window
164K tokens
Max output
66K tokens
Input $ / 1M tokens
$0.28
Output $ / 1M tokens
$0.42
Cached input $ / 1M
$0.028
Each benchmark shown separately with its own source

Benchmark results (6)

BenchmarkDomainScorevs best recordedSettingRunSource
LMCAagents15.2%
22%
——External ↗
DTBenchreasoning62.7%
63%
——External ↗
Chess Puzzlesgames14.0% ±3.5
19%
—16 Dec 2025Eval log ↗
Aider polyglotcoding74.2%
84%
——External ↗
OTIS Mock AIME 2024-2025math87.8% ±4.1
88%
—16 Dec 2025Eval log ↗
GPQA diamondscience83.4% ±2.0
87%
—16 Dec 2025Eval log ↗

Source: Epoch AI Benchmarking Hub (CC BY 4.0). “External” rows are leaderboard results Epoch collects from third parties. Best reported setting per benchmark is shown.

API list price over time

Price history

$ per 1M tokens

12 recorded prices on 6 Dec 2025 (OpenRouter listing + Internet Archive snapshots); re-read on every data refresh, most recently 8h ago. Steps show when the price changed.

Events