Skip to content
DeepSeek

DeepSeek v3

Maker’s page
Check price at Amazon for DeepSeek DeepSeek-V3 (affiliate link, opens in new tab)
Weights at 16-bit
1,342 GB
Weights at 8-bit (Q8_0)
713 GB
Weights at 4-bit (Q4_K_M)
407 GB
Smallest card at 4-bit
More than one card
Parameters
Active parameters
Layers
KV heads
Context
Licence

Specs

Active parameters
Attention heads
Context
Downloads (30 days)
Experts
Experts per token
Family
GGUF downloads
repobytesfilesquant
unsloth/DeepSeek-V3-GGUF221,253,686,3685 entriesQ2_K_XS
unsloth/DeepSeek-V3-GGUF244,245,533,7925 entriesQ2_K_L
unsloth/DeepSeek-V3-GGUF319,198,395,2967 entriesQ3_K_M
unsloth/DeepSeek-V3-GGUF404,430,187,0729 entriesQ4_K_M
unsloth/DeepSeek-V3-GGUF475,396,557,60010 entriesQ5_K_M
unsloth/DeepSeek-V3-GGUF550,798,326,36812 entriesQ6_K
unsloth/DeepSeek-V3-GGUF713,286,514,08016 entriesQ8_0
Hidden size
KV heads
KV latent rank (MLA)
Layers
Licence
Parameters
Parameters by type
Publisher
Released
RoPE key size (MLA)
Shared experts
Task

More from DeepSeek

Across the range 7 models by family