Reaction
Concern over Moonshot’s potential K2.8 inference costs
A poster worries that the “lightweight” K2.8 could be expensive to run if it resembles K2+.
TLDR
A poster argues that if Moonshot’s “lightweight” K2.8 is another K2+, it would be far more expensive at inference than any DeepSeek model since April. They worry Moonshot may have scaled too early.
Combined views
12.8K
1 Source, first seen ago