BREAKING
DeepSeek Releases Open-Weight V4 Models
0
T
V4-Pro total params
0
B
V4-Flash total params
0
M
token context window
How Microsoft Is Approaching V4
1
Host on Azure Foundry
↓
2
Weigh secured deployment
↓
3
Keep data out of China
↓
4
Target Copilot Cowork
V4-Pro Benchmark Scores
MMLU
90.1
HumanEval
76.8
MMLU-Pro
73.5
Cost Edge vs Control Concerns
Strengths
Low cost
●
~$1.74-$1.93 per M tokens
●
Strong coding and agentic work
●
One million token context
Concerns
Caution
●
Trails top U.S. proprietary models
●
Security risk from Chinese origin
●
Needs data isolation
Cost Edge vs Data Control
AI NEWS BLITZ
DeepSeek has open-sourced its V4 series, drawing interest from Microsoft while still trailing top U.S. systems.