A large meta-analysis from 2023 (178 studies) shows that higher loads yield the greatest gains for strength, while muscle mass grows at virtually any weight level as long as total effort is high enough. Two to three training sessions per week with multiple sets and a challenging weight is the proven starting point. The difference between a good and an optimal program is small compared to the difference between training consistently and not training at all.
Training two to three times per week, doing multiple sets per exercise, with a weight that genuinely challenges you by the end of a set: that is the starting point the evidence points to. A large 2023 analysis of 178 studies compared virtually every conceivable training schedule and found that higher loads (roughly the heaviest weight with which you can still perform 6 to 8 repetitions with correct technique) produce the greatest gains in strength, while muscle size grows at almost any weight level as long as the total effort is kept high enough.
For strength and muscle mass, the exact weight therefore matters somewhat more than it does for volume alone. But the difference between a 'good' and a 'very good' programme is small compared with the difference between training consistently and not training at all. Technique and gradual progression over weeks and months carry at least as much weight as the precise load.
If you are somewhat older, the same principle applies, but with one nuance: strength and functional ability (climbing stairs, rising from a chair) improve noticeably, while the visible increase in muscle size is somewhat more modest. That is no reason to approach things differently, but it is a reason to use strength and ease of movement as your benchmark rather than the mirror alone. Resistance bands at home or weights at a gym both work; if you have heart or joint problems, it is wise to first consult a doctor or physiotherapist about how to build up gradually.
Overview across multiple factors (2 research records, 4 sources). The strength of evidence varies by component -- read the answer for the nuance.