SITUATION EXPLAINED: DeepSeek V4.1 Flash beats its own Pro model from 3 weeks ago. • 552B mixture of experts, 8B active for input and 16B for output. MIT licensed, 1M context • Beats V4 Pro on every benchmark except GPQA Diamond and Humanity's Last Exam • $0.15 input, $0.60 output off-peak. Cache hits are $0.003 • Terminal-Bench 3.0 went 11.8 to 30.0. DeepSWE went 62.7 to 74.2. ExploitGym went 5.4 to 15.3 • Leads Sol and Opus 5 on DeepSWE, CyberGym, AutomationBench, and Agent's Last Exam.…

DeepSeek@deepseek_ai14h ago

🚀 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