Is GPT-5 pricing confirmed?

No. This page is a planning aid, not a price list. Treat any figure here as a scenario, and re-model with published rates once they exist.

How should I budget for a new model tier?

Model your current workload at your existing rate, then run scenarios at a higher and lower rate. The spread tells you how exposed your budget is to a pricing change.

Should I wait before migrating?

Migrate when a modelled cost or quality gain justifies the work, not on announcement. Keep the current scenario saved so the comparison is concrete.

OpenAI GPT-5 Pricing: What to Expect and Plan For

What pricing model to expect

OpenAI is converging on a reasoning-token model: visible output tokens billed at one rate, hidden reasoning tokens billed alongside. Plan for $15+/1M output in reasoning modes.

Reference points from o-series

o1: $15 in / $60 out per 1M. o3-mini: $1.10 / $4.40. GPT-5 is likely positioned above o1 on capability, with a similar or higher rate card.

Budgeting before launch

Build cost models assuming 2–3× the GPT-4o rate for the flagship tier and a mini variant at GPT-4o-mini parity. Most teams overshoot by assuming everything must run on the flagship.

Migration cost planning

Plan a routed rollout: 5–10% of traffic on the new tier for eval, then graduate by intent. Direct migrations from GPT-4o to a frontier reasoning model can 10× the bill overnight.

Stay current

GPT-5 is not in the TokenAtlas model catalog, because OpenAI has not published a rate card for it. When a rate card is published, the model can be added to the catalog and you can re-run your workload against it in the calculator. Until then, model it with the o1/o3 reference rates above.

How to model a hypothetical GPT-5 migration

Since GPT-5 has no published SKU here, treat any figure as a planning assumption, not a quote. As one hypothetical: apply o1's $15/$60 rate to your current GPT-4o call volume (100K calls, 4K in/600 out) — input 400M*$15/1M=$6,000, output 60M*$60/1M=$3,600, total $9,600 versus $1,600 on GPT-4o today, a 6x hypothetical increase. This is a scenario to stress-test budgets, not a forecast.

Mechanics: what a reasoning-token component changes

Reasoning models like o1 generate internal reasoning tokens that are billed as output but not shown to the user, so the visible answer length understates the actual output token count. Budgeting from visible response length alone on a reasoning-tier model will systematically undercount cost; historical o1 usage suggests hidden reasoning tokens can multiply effective output volume several times over for complex prompts.

Decision checklist before committing budget to an unreleased tier

— Have you modelled cost at 2-3x today's flagship rate as a stress case? — Is there a mini/nano equivalent tier likely to launch alongside it? — Can your routing layer fall back to GPT-4o if the new tier's cost exceeds a threshold? — Have you set a hard spend cap for the eval rollout period?

What this post deliberately does not do

This post does not quote a GPT-5 price because none exists publicly at time of writing — any number attached to GPT-5 here is explicitly a hypothetical built from o1/o3 reference rates, not a forecast of OpenAI's actual pricing decision. Treat every figure in this piece as a planning input to pressure-test your budget, then replace it with the real rate card the day it's published.

Frequently asked questions

Is GPT-5 pricing confirmed?: No. This page is a planning aid, not a price list. Treat any figure here as a scenario, and re-model with published rates once they exist.
How should I budget for a new model tier?: Model your current workload at your existing rate, then run scenarios at a higher and lower rate. The spread tells you how exposed your budget is to a pricing change.
Should I wait before migrating?: Migrate when a modelled cost or quality gain justifies the work, not on announcement. Keep the current scenario saved so the comparison is concrete.