[Official FAQ] Corrections and follow-up questions welcome.

Phi-class models target strong results at small sizes—handy on laptops. Yi and GLM families are often discussed for bilingual/Chinese-quality chat among other strengths. Treat this as a map, not a ranking.

Shortlist by constraint:

• Edge / low VRAM: try small Phi-class Instruct builds.
• Same VRAM budget: A/B Yi, GLM, Qwen, Llama Instruct siblings on your prompts.
• Filter by license, context length, and available GGUF/official weights.

Small models bluff on multi-step tasks—keep a tiny regression set. Comparative posts with fixed prompts help everyone.