Best LLMs for your hardware
Pick a build you own (or want to buy). We surface the open-weights LLMs that fit at your chosen quant, ranked across four categories — and put the closed-frontier four (Gemini 3.1 Pro, GPT-5.5, Sonnet 5, Opus 4.7) next to them so you can see exactly how far behind local actually is in May 2026.
Shopping for one specific model? See Qwen 3.6 27B VRAM & GPU requirements — per-quant VRAM, a "can my GPU run it" table and measured tokens per second for every card that fits it. Weighing a unified-memory box? RTX Spark vs DGX Spark vs RTX 5090 maps capacity vs speed across all three. On the software side, CUDA vs Vulkan for llama.cpp explains which backend to actually run on each vendor's card.
New to all this? Start with How to Run AI Locally: the complete beginner's guide — run your first model in 10 minutes, then pick hardware once you know what you need.
Pick a build above to see which open-weights LLMs fit and how they stack up against closed frontier models.
Closed-frontier reference (May 2026)
What "API-grade" actually scores right now. Use this as the ceiling — anything local will lag here by some margin, and that's fine for most workflows.
Every open-weights model that fits, ranked by composite score
Composite blends benchmark averages (60 %) with editorial 0-5 ratings (40 %). Click any column header to sort — closed-frontier references mix into the ranking and stay amber-tinted wherever they land.