DeepSeek v4.1-Flash multimodal model lands on Modal with 16B/8B asymmetric activation

charles_irl · x · 2026-09-12

DeepSeek's new multimodal model v4.1-Flash is now available on Modal. It uses DeepSeek's Causal Encoder-Decoder architecture, activating 16B parameters for input and 8B for output to cut costs on input-heavy agentic workloads.

Original post →

More from coding & agent

coding & agent channel →