Model Deployment

This is the largest cluster on the blog, and the most literal: step-by-step guides for getting a specific open-source model running on a GPU you rent. It covers text models (Qwen, Gemma, Granite, GPT-OSS, Ministral, Phi-5, Command A, Nemotron), vision-language and image models (SAM 3, Ideogram 4, FLUX.2, Qwen-Image-Edit), video and audio (Wan, SmolVLM, faster-whisper, open-source TTS), and infrastructure that sits around a model rather than being one: vector databases, embedding servers, RAG pipelines, AI gateways, and guardrails.

Every post here follows roughly the same shape: exact VRAM requirements, the deployment command (usually vLLM, sometimes SGLang or a model-specific runtime), and a real cost figure on Spheron hardware, because "how much GPU do I need for this" is the question that stalls most deployments before they start.

The pillar guide, GPU Requirements Cheat Sheet 2026, is the fast-reference version: one table with VRAM, the cheapest working GPU configuration, and live hourly cost for eighteen models, updated as new releases ship. Use it to get a rough number in seconds, then open the specific model's post in this cluster for the full setup.