Saltar al contenido
EN ES
DeepSeek

DeepSeek V3.2 Speciale

Released 2025-12 · reasoning · 164K tokens · 3 benchmarks · Open weight
HuggingFaceGitHubPaper

Editorial notes

Variante de razonamiento profundo, 685B MoE. Medallas de oro en IMO 2025 e IOI 2025. Paper oficial reporta solo 4 benchmarks "razonamiento puro" (GPQA / AIME / LiveCodeBench / HMMT 99.2) con scaled test-time compute. MMLU-Pro / SWE-bench / Aider no se publicaron para Speciale.

Spec sheet

Empresa
DeepSeek
Pais
CN
Tipo
reasoning
Release
2025-12
Context
164K tokens
Params total
685B
Params active (MoE)
37B
Licencia
deepseek-license
Quants
BF16, Q8_0, Q5_K_M, Q4_K_M, Q3_K_M
Pricing (deepseek)
$0.27/$1.1/M
Slug
deepseek-v3-2-speciale

Quick install

2 tools
ollama.com 
ollama run deepseek-v3.2:speciale

Benchmarks (3)

Reasoning 1

Math 1

Coding 1

Cite this model
BibTeX · APA

BibTeX 

@misc{frontier-deepseek-v3-2-speciale,
  title  = {DeepSeek V3.2 Speciale},
  author = {{DeepSeek}},
  year   = {2025},
  note   = {Frontier Benchmarks AI atlas. Accessed 2026-05-08},
  url    = {https://frontierbenchmarks.com/models/deepseek-v3-2-speciale}
}

APA

DeepSeek (2025). DeepSeek V3.2 Speciale [Large language model]. Frontier Benchmarks AI. Retrieved 2026-05-08, from https://frontierbenchmarks.com/models/deepseek-v3-2-speciale

Citation refleja la pagina del atlas, no el paper original del modelo. Para el paper, ve a la seccion "Recursos" arriba.

⚔️ Battle vs otro modelo ← Todos los modelos Mas de DeepSeek Methodology