DeepSeek-V4-Flash-Vision-Exp on local hardware
Estimated comfort for repository coding with Aider on a MacBook Pro M4 Pro 48GB.
DeepSeek-V4-Flash-Vision-ExpOpen source
deepseek-ai · mit · 476,654 downloads
Mixture of expertsVision
- Parameters
- 305B exact
- Active per token
- ≈34B (6 of 256 experts)
- Context window
- 64K tokensRoPE-extended to 1M (quality degrades past native — treat long-context ratings as optimistic)
- Attention
- Sliding-window + global layers
- Layers · KV heads · head dim
- 43 · 1 · 512
- KV cache
- 0 KB/token · 0.0 GB at 32K
- Tool calling
- basic (inferred — no supporting signal)
- Thinking
- basic (inferred — no supporting signal)
- Vision
- good (from config)
- Commercial use
- allowedRetain the required license and attribution notices
- Import confidence
- medium · capabilities estimated
Weights by quantization
Q3_K_M
119 GB
gguf file size
Q4_K_M
144 GB
gguf file size
MXFP4
144 GB
gguf file size
- • Uses a compressed / sparse attention scheme; KV-cache size is approximate.
- • Context native 65536, rope-extended to 1048576 (yarn); quality degrades past native — treat long-context ratings as optimistic.
- • No chat template found; tool calling is inferred as limited (low confidence).
- • toolCalling: basic, inferred from missing signals (low confidence).
- • thinking: basic, inferred from missing signals (low confidence).
- • vision: good, inferred from config (low confidence).
- • Capability ratings are estimated from model size and name — this model hasn't been assessed by hand.
- • GGUF sizes come from community conversions matched by model name; equivalence and runtime support are not verified. MLX conversion availability is unverified.
DeepSeek-V4-Flash-Vision-Exp — 3-bit
Does not fit in memory
Needs ≈127 GB but only ≈37 GB is available after the OS, other apps and Aider.
High confidence