DeepSeek V3 (Dec '24)
DeepSeekDeepSeekOpen WeightMIT + Model License (Commercial use allowed)
Descripción
A powerful Mixture-of-Experts (MoE) language model with 671B total parameters (37B activated per token). Features Multi-head Latent Attention (MLA), auxiliary-loss-free load balancing, and multi-token prediction training. Pre-trained on 14.8T tokens with strong performance in reasoning, math, and code tasks.
Fecha de lanzamiento
2024-12-26
Parámetros
671.0B
Longitud del contexto
—
Modalidades
text
Radar de capacidades
28
general
29
coding
38
reasoning
38
science
35
agents
0
multimodal
Rankings
| Dominio | #Posición | Puntuación | Fuente |
|---|---|---|---|
| Ranking de codificación | 441 | 28.0 | AA |
| Ranking general | 434 | 32.0 | AA |
| Ciencia | 426 | 35.0 | AA |
Puntuaciones de benchmarks (LLM Stats)
(LLM Stats (zeroeval))Chat
IFEvalGoogle Research (2023)
86.1%Aut.
Factuality
SimpleQA
24.9%Aut.
General
MMLU
88.5%Aut.
C-Eval
86.5%Aut.
Aider-Polyglot Edit
79.7%Aut.
CSimpleQA
64.8%Aut.
Aider-Polyglot
49.6%Aut.
Language
CLUEWSC
90.9%Aut.
MMLU-Redux
89.1%Aut.
MMLU-Pro
75.9%Aut.
Long Context
LongBench v2
48.7%Aut.
Math
MATH-500
90.2%Aut.
CNMO 2024
43.2%Aut.
AIME 2024
39.2%Aut.
Reasoning
DROP
91.6%Aut.
HumanEval-Mul
82.6%Aut.
FRAMES
73.3%Aut.
GPQANYU + Cohere + Anthropic (2023)
59.1%Aut.
SWE-Bench Verified
42.0%Aut.
LiveCodeBench
37.6%Aut.
Índices de evaluación AA
(Artificial Analysis)Math 500(OpenAI (2024), subset of Hendrycks et al. MATH (2021))88.7
Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))75.2
Gpqa(NYU + Cohere + Anthropic (2023))55.7
Livecodebench(UC Berkeley + MIT + Cornell (2024))35.9
Scicode(UIUC + Argonne National Lab (2024))35.8
Ifbench(Google Research (2023))34.8
Lcr(Artificial Analysis)29.3
Math Index(Artificial Analysis)26.0
Aime 25(MAA (Mathematical Association of America))26.0
Aime(MAA (Mathematical Association of America))25.3
Coding Index(Artificial Analysis)23.0
Tau2(Sierra + U Toronto + Vector Institute (2025))22.8
Terminalbench V2 116.9
Intelligence Index(Artificial Analysis)8.5
Terminalbench Hard(Stanford × Laude Institute (2026))6.8
Tau Banking4.7
Hle(Center for AI Safety + Scale AI (2025))2.9
Terminalbench V4 00.0
Puntuaciones por categoría LLM Stats
(LLM Stats (zeroeval))Chat90
Instruction Following90
Language80
Legal80
Finance80
Healthcare80
Math70
Reasoning70
Search70
Structured Output70
General70
Physics60
Biology60
Chemistry60
Long Context50
Code50
Frontend Development40
Factuality20
Precios
Precio de entrada$0.32 / 1M tokens
Precio de salida$0.89 / 1M tokens
Precio mixto (3:1)$0.463 / 1M tokens
Velocidad
Tokens/seg0.0
Retraso del primer token0.00s
Tiempo hasta la respuesta0.00s
Ranking de Precios por Proveedor
Ranking de Precios por Proveedor
4 proveedores
Más barato: DeepInfraMás caro: Helicone
ProveedorEntradaSalida
1DeepInfraMás barato
$0
$0
2Alibaba (China)
$0.287
$1.147
3DeepSeekPRINCIPAL
$0.32
$0.89
4Helicone
$0.56
$1.68
Comparar precios entre diferentes proveedores de API para este modelo.