The broadest-default frontier model. Strong tool ecosystem, cheapest mid-tier in the cohort (5-mini, 5-nano), and the most third-party SDK support. Loses to Opus 4.7 on long-context agent reliability and to Gemini on video; wins on price-performance for everything else.
GPT-5 is the broadest-default frontier model — strong general performance, sharp tier pricing, and the deepest third-party SDK ecosystem.
What the three tiers buy you
Flagship 5 handles frontier-grade tasks at frontier prices. 5-mini is the new “good enough” for most production workloads at a fraction of the cost. 5-nano covers the high-volume cheap-call lane where you’d previously have reached for an open-weights model.
Where it loses
Long-context agent reliability — Opus 4.7 still feels more disciplined in 30+ minute autonomous runs. Video understanding — Gemini 2.5 Pro is sharper. On cost-conscious general-purpose work, GPT-5 wins.