· 3 min read
Edition 012 — DeepSeek shrinks the KV cache, SWE-2 pushes coding's Pareto frontier, and VLMs learn to play robots
DeepSeek V4.1 Flash compresses the global KV cache to 890 bytes/token, Cognition's SWE-2 pushes the Pareto frontier of agentic coding, and two research papers show VLMs playing robots and multi-agent pipelines optimizing themselves 2.5x faster.