Price watcher · unreachable

Checked 18 Sept 2026. HTTP 403

ChatGPT Plus

Like other first-party assistant subscriptions, a daily user's conversations are estimated to cost more through the published API than the subscription charges.

FAIR

The price is mostly buying things that are not the model: evals, data pipelines, integrations, compliance, support, real interface work.

Vendor
OpenAI
Category
Assistant
Tier
Plus
Price / month
$20
Est. API cost
$40.50
Est. markup
0.49×
Price checked
13 Aug 2026
Verification
verified
§1 · Billing

What the price actually is

Billed monthly. Higher Pro and Business tiers exist and are not covered by this entry. OpenAI blocks automated requests to its pricing pages, so this figure cannot be machine-checked and is confirmed by hand.

§2 · Assumption

The usage this is priced against

A daily user with 200 conversations a month, each carrying roughly 10,000 tokens of accumulated context and returning 700 tokens, answered by a large frontier model.

Someone using it half as much sees double the multiple. The assumption is the argument — if you disagree with it, the number below is not about you. Change it in the calculator.

§3 · Arithmetic

The cost math, in full

Estimated monthly inference cost

2,000,000 input tokens$30.00

Large frontier text model at $15.00 / 1M input tokens · 200 conversations at typical consumer thread length

140,000 output tokens$10.50

Large frontier text model at $75.00 / 1M output tokens

Estimated cost$40.50
Price paid (Plus)$20
Estimated markup0.49×

Rates come from a dated rate card of representative published API prices, not from the vendor. Nobody outside these companies knows what they actually pay; volume discounts and in-house serving both push real costs below these figures, which makes every multiple here a floor rather than a ceiling.

§4 · Reasoning

Why this verdict, not the number

This is the product most people mean when they say they pay for AI, and it is not a wrapper around anything — it is the model vendor selling access to its own model. The flat fee sits below the estimated API cost of the assumed usage, which is what a subsidised consumer tier looks like. Bundled tools and image generation are not priced in here at all, so the real gap is wider than shown.

What the price buys besides tokens

  • Proprietary model
  • Scale & infra
  • Distribution

In-house models plus the serving capacity to answer at consumer scale, and a client most people can use without instruction. The bundled extras — browsing, image generation, data analysis, voice — are additional inference this estimate does not attempt to price.

§5 · Leaving

What you lose if you leave

The specific models and everything bundled around them. There is no cheaper route to the same product, only routes to different products.

The honest cheaper path

The API with your own client, which is cheaper for light use and more expensive at the volume assumed here. Open-weight models running locally are free of marginal cost and are not the same thing.

§6 · Your numbers

Recompute it for yourself

Recompute this for your own usage

Your estimated cost
Your markup
Cost per unit of our assumption

Scaling assumes your usage has the same shape as ours, just more or less of it. If your mix is different — far more output than input, say — the estimate drifts. It is an estimate either way.

§7 · Sources

Where every number came from

Price recorded 13 Aug 2026 · entry last reviewed 13 Aug 2026. Think something here is wrong? File a correction — we publish them, including the ones that embarrass us.