August 2026 Summaries

Vast.ai announced expanded availability of NVIDIA B200 GPUs through August and September, enabling hourly rentals and longer-term reservations for AI training and production inference, alongside support for ARM-based NVIDIA GPUs such as GB10 systems. Its Model Library has added Kimi K3, a 2.8-trillion-parameter multimodal mixture-of-experts model with a one-million-token context window, available in full-precision and quantized deployment options, as well as templates for models including Kimi K2.7 Code, Gemma 4 variants, DeepSeek V4 Flash, Hunyuan HY3, and Krea 2 Turbo. Platform updates include configurable renter notifications for lifecycle events, outbids, and disk warnings, HTTPS webhooks with programmatic API, CLI, and SDK management, and direct Hugging Face Storage Bucket connections for transferring datasets, checkpoints, and results without manual uploads. Additional changes improve billing visibility, serverless analytics, search and template-editor interfaces, while resolving issues involving template filters, earnings displays and exports, email delivery timing, and page crashes.