deepseek v4.1 flash is kind of crazy in 1 month they managed a new architecture, a new pretrain, and significantly scaled RL training again... 1 month! the arch is great for efficiency, but what's significant to me is the post-training scaling. most of the improvement is essentially just more compute + more data at the RL stage at this point the model is mostly made in post-training

DeepSeek@deepseek_aiSep 10

🚀 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