Mistral Small 4 Achieves GPT-4-Class Performance—But at What Hardware Cost?

Mistral Small 4 delivers GPT-4-level coding and reasoning with open weights, but the 74 GB footprint makes local deployment economically impractical—requiring $3K–4K in hardware versus a $0.15/M-token API. The real story is how open-source advantage increasingly depends on infrastructure economics rather than capability alone.