DeepSeek always manages to surprise the industry with its releases. DeepSeek-V4.1-Flash is a new small model that tops GPT 5.6 Sol and Opus 5 in CyberGym and DeepSWE 1.1. It is a 552B-parameter MoE with a Causal Encoder–Decoder architecture. It has 8B active parameters for input and 16B for output. The price are insanely too. You get to spend $0.15 in off-peak hours for input and $0.6 input for output (1m tokens). Those raise to $0.3 and $1.2 in peak hours.
🚀 Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient.
🔹 Introducing the smallest model in our new architecture family, with native visual understanding.
🔹 Designed for greater capability, faster inference, higher throughput, and scaling to larger models.1/6 pic.twitter.com/wxJGiyX56o
— DeepSeek (@deepseek_ai) September 10, 2026
DeepSeek 4.1-Flash is now live on the API. With V4.1 Flash, DeepSeek can serve more users at a lower price.
[HT]
