Compare AI models

Specifications
Parameters
671B671B
Context window
128k128k
API pricing
Cheapest input
$0.50DeepInfra—
Cheapest output
$2.15DeepInfra—
$208DeepInfra · fp4—
Benchmarks
LiveCodeBench
73.3%88.7%
Humanity's Last Exam · text only, no tools
17.7%30.6%
AIME 2025
87.5%96%
GPQA Diamond
81%85.7%
Overview
CompanyDeepSeekDeepSeek
Release dateMay 28 2025Dec 1 2025
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

DeepSeek-V3.2-Speciale leads DeepSeek R1-0528 on 4 of the 4 benchmarks they both report (LiveCodeBench, Humanity's Last Exam, AIME 2025, GPQA Diamond). DeepSeek R1-0528 shipped 187 days before DeepSeek-V3.2-Speciale, so benchmark comparisons should account for the intervening progress.

DeepSeek R1-0528 has 671B parameters, while DeepSeek-V3.2-Speciale has 671B. Context windows are 128k (DeepSeek R1-0528) vs 128k (DeepSeek-V3.2-Speciale).

On LiveCodeBench, DeepSeek-V3.2-Speciale leads at 88.7% vs DeepSeek R1-0528 at 73.3%. On Humanity's Last Exam · text only, no tools, DeepSeek-V3.2-Speciale leads at 30.6% vs DeepSeek R1-0528 at 17.7%. On AIME 2025, DeepSeek-V3.2-Speciale leads at 96% vs DeepSeek R1-0528 at 87.5%. On GPQA Diamond, DeepSeek-V3.2-Speciale leads at 85.7% vs DeepSeek R1-0528 at 81%.