BREAKING
DeepSeek-V4-Flash Runs Locally
0B
total params
0B
active params
0M
context tokens
Memory by quant variant
1-bit92
2-bit102
3-bit135
4-bit162
Q8169
V4-Flash benchmark scores
MMLU-Pro86.2
GPQA Diamond88.1
Terminal Bench 2.056.9
Local run: gains and cautions
Upsidespraised
Lossless Q8 near 168GB
3-bit works on 110GB Mac
Tool-use and chat fixes
Cautionsnoted
Q2/Q3 degrade long tasks
Cloud version still ahead
Large download times
Open weights, now on your desk
AI NEWS BLITZ
Unsloth has released quantized GGUFs that let DeepSeek-V4-Flash run fully offline.