Buoyant has released Kinference, an Apache 2.0 inference control plane for vLLM on Kubernetes that adds declarative routing policies, KV-cache-aware load balancing, and per-tenant metrics. In Buoyant’s own benchmark, the system cut median time-to-first-token by 95 percent. (MarkTechPost)
Recursive Intelligence is releasing Ripple-65B Preview, a 65-billion-parameter open-weight model fully licensed under Apache 2.0 that achieves the top score among open-source U.S. models on TerminalBench 2.1 for agentic coding at a total R&D cost of $32 million. (The Decoder) Separately, Nous Research retrained Alibaba’s Qwen 3.8-27B into Arete-27B using 53 million tokens of supervised fine-tuning, reducing the model’s propaganda rate from 22 percent to less than one percent while preserving general capabilities. (The Decoder)