Deepseek open-sourced DeepInfer, the vLLM-based inference framework behind its V4.1-Flash model, aiming to reduce compute costs and energy consumption when running AI models at scale (The Decoder). The release adds pricing pressure on Western providers by making efficient inference infrastructure available to developers building self-hosted systems.
Google Cloud, alongside Red Hat and IBM, released llm-d 1.0, an open-source Kubernetes-native stack for distributed LLM inference that adds multi-tenant flow control, zero-downtime upgrades, and agent-aware routing across vLLM, SGLang, TPUs, and GPUs (MarkTechPost). Separately, Mistral AI launched Voxtral 3, a family of speech-to-text and text-to-speech models paired with a LiveKit-based voice agent SDK that can run on self-hosted infrastructure and supports real-time tool integration across more than 50 languages (The Decoder), while Elevenlabs released two new speech-to-speech models with 98.8 percent accuracy on Big Bench Audio and 0.57-second response time (The Decoder).