Model upgrades may now be more about cost than magic
Ars Technica
26 July 2026
How this applies to you
If you are paying for AI across a team, test cheaper models on real tasks before upgrading everyone. You may find that drafting, summarising and classification work fine on lower-cost options.
Ars Technica reports that Anthropic's Opus 5 is focused on token efficiency rather than a large capability jump. The point for buyers is simple: the best model is not always the most expensive one.
Why it matters. If you are paying for AI across a team, test cheaper models on real tasks before upgrading everyone. You may find that drafting, summarising and classification work fine on lower-cost options.
