BREAKING
DeepSeek Unveils V4-Pro Model
0
T
total params
0
B
active per token
0
M
context window
80.6% on SWE-Bench Verified
SWE-Bench
80.6
US lead (months)
8
Value vs Full Parity
Strengths
Value
●
MIT open-weight license
●
API 1/5 to 1/10 of Western cost
●
Strong long-context handling
Concerns
Gaps
●
Beaten by GLM and Kimi on some tests
●
Preview instability, peak latency
●
Relies on domestic Huawei chips
0
B
raised
0
B
valuation
Three Races At Once
AI NEWS BLITZ
DeepSeek is chasing US AI rivals with new models, big funding, and its own chips.