GPT-5 review: GPT-5-Thinking is a substantial upgrade over o3-pro, Auto is only useful for free tier users, picking the right model still matters, and more
What do I ultimately make of all the new versions of GPT-5? — The practical offerings and how they interact continues to change by the day.
Context & Ripple Effects
GPT-5 arrived framed as a unified system: an efficient model for routine questions, a reasoning model for harder work, and a router intended to choose between them. This review finds that the promise of automatic routing does not remove the need for informed selection, particularly outside the free tier.
The assessment also extends a recent arc in which GPT-5 was seen as competent but not a uniform step-change against rivals. Its strongest distinction here is a more capable Thinking option, rather than a single model that is best for every task.
First-order effects
- Users who need demanding reasoning have a clearer reason to choose GPT-5-Thinking over o3-pro, while users seeking automatic selection get limited value from Auto beyond the free tier.
- Model choice remains an active operating decision: users must match task requirements to the practical GPT-5 offering rather than treat the product family as interchangeable.
Second-order effects
- OpenAI’s unified-system positioning is tested by the usability of its routing layer: if capable users still choose manually, product differentiation shifts toward clear model tiers and guidance rather than automation alone.
- For buyers comparing providers, aggressively competitive GPT-5 pricing matters less than whether a chosen model reliably delivers the needed reasoning quality—an instance of performance being judged against competing LLMs and task fit.
Third-order effects
- AI product competition may increasingly center on reasoning economics: providers will need to package stronger reasoning models, lower-cost defaults, and routing in ways that make the quality-cost tradeoff legible.
- If model portfolios continue to change quickly, organizations will need repeatable evaluation and governance practices rather than one-time model standardization; automatic routing may help, but it does not yet eliminate selection risk.
The trend: Frontier AI is moving from single-model launches toward tiered, routed model portfolios where the practical value depends on matching reasoning capability and cost to each task.