This browser has no WebGPU, so a model cannot run here. Chrome or Edge on a machine with a GPU will work — or pick a local Ollama model instead.

• Llama 3.2 · 1B~0.9 GBSmall, fast and surprisingly useful. The sensible default.Download
• Llama 3.2 · 3B~2.1 GBNoticeably better reasoning, needs a real GPU.Download
• Qwen 2.5 · 1.5B~1.1 GBStrong at code and structured output.Download
• Qwen 2.5 Coder · 1.5B~1.1 GBA code-tuned model. Good at reading a diff.Download
• Phi 3.5 mini~2.2 GBPunches above its weight, especially on reasoning.Download
• Gemma 2 · 2B~1.9 GBA good all-rounder from Google.Download
• SmolLM2 · 1.7B~1.1 GBThe lightest useful option — a good first download.Download