How these rankings work
Each guide in this hub picks winners for a single, narrow use case rather than trying to name one best model or one best tool overall — "best coding model on a 24GB GPU" and "best model for creative writing" have different right answers, and a ranking that pretends otherwise is less useful than eight separate ones. Every pick links through to the model, tool, or hardware benchmark page it draws its numbers from, so a claim on a "best of" page is never the only place that number appears on the site.
Picks are re-tested rather than re-typed when a new model or tool release changes the picture. Qwen, DeepSeek, GLM, and Llama have all shipped multiple point releases through 2026, and a "best" list that isn't updated against them stops being a ranking and becomes a historical snapshot — that's the reason this hub is dated per page and revisited monthly rather than left static after publication.
What the eight guides cover
Three guides are model rankings at different scopes: an overall 24GB-GPU leaderboard, a coding-specific shortlist for the same hardware class, and a creative-writing and roleplay list that weighs frontend software as heavily as the model itself. Two are hardware guides — a from-scratch RTX 3090 build and a comparison across the current Apple Silicon Mac Studio lineup — for readers deciding what to buy before they've picked a model at all. The remaining three cover software categories that sit underneath every model choice: inference engines (how a model actually gets served), fine-tuning frameworks (how you'd adapt one to your own data), and Stable Diffusion UIs for image rather than text generation.
If you're starting from zero, the build guide and the inference-engine guide are the right entry points — hardware and serving software are decisions you make once, while the model rankings are the ones worth revisiting as new releases land. If you already own a GPU or Mac, jump straight to the model or use-case guide that matches your hardware class.
Reading a pick against the rest of the site
A "best" designation on this hub is a shortcut to a more detailed answer that lives elsewhere on the site, not the final word on its own. Every model named in these guides has its own model page with full VRAM figures across sizes, and every GPU or Mac referenced has its own hardware benchmark page with bandwidth, price, and generation context. If a pick here seems to conflict with something more specific on a model or hardware page — a newer point release, a corrected VRAM figure — the more specific page is the one to trust; this hub is a starting point for narrowing choices, and the detail pages are where the underlying numbers actually live.