Chinese AI laboratory DeepSeek released the V4-Flash-High model, which scored 1586 on arena.ai’s Frontend Code Arena, ranking 7th overall and 3rd among open-weight models. This performance represents a 154-point jump from the previous V4-Pro-Preview version. The model uses a Mixture-of-Experts architecture with 284 billion total parameters and 13 billion active parameters, priced at 0.14 dollars per million input tokens and 0.28 dollars per million output tokens, with a 1 million token context window. This breakthrough validates the thesis that AI inference can be commoditized, a market specifically targeted by decentralized compute protocols like Akash, Render, and io.net.
Source: Read the original article
