OverpayingForAIPricing desk

Lesson 2 of 8 · 10 min read · Beginner

OpenAI's models explained: which GPT do you actually need?

GPT-5.5, GPT-5.4, mini, nano, Pro and gpt-oss all exist for a reason. Learn what each tier is for, how much it costs per million tokens, and when a cheaper model is the smarter choice.

In this lesson you will

  • Map OpenAI's current model tiers (flagship, standard, mini, nano, Pro, open-weight) to the jobs they suit
  • Read a per-token price and estimate what a task will cost
  • Explain what a context window is and why a bigger one is not automatically better

OpenAI sells a ladder of models, not one model. In ChatGPT the picker shows a handful of options and the app often auto-routes for you. In the API you choose explicitly and pay per token. Either way, the choice matters: the gap between the cheapest and the most expensive model in the current line-up is well over a hundred times on output price.

OpenAI API models, $ per 1M tokensLive from the OverpayingForAI catalogue · last verified 2026-09-04 · sorted by output priceInputOutputGPT-5.51.1M context$5$30GPT-5.41.1M context$2.5$15GPT-5.6 Terra1.1M context$2$12GPT-5.6 Sol1.1M context$2$10GPT-4.11.0M context$2$8GPT-5.4 mini400K context$0.75$4.5GPT-5 Mini400K context$0.25$2GPT-4.1 Mini1.0M context$0.4$1.6Batch, fast-mode and free variants are excluded. Prices change; the figure re-draws from the catalogue on every build.
Figure 1.Live from our catalogue: input and output price per million tokens for OpenAI's current text models. Output tokens are always the expensive side, which is why long answers cost more than long questions.

The tiers, in plain language

  • Flagship (GPT-5.5) — the strongest general model. Use it for hard reasoning, long multi-step work, and anything where a wrong answer is expensive.
  • Standard (GPT-5.4, GPT-5.6 Sol / Terra, GPT-4.1) — the daily driver. Good enough for most writing, analysis and coding at a fraction of flagship output price.
  • Mini (GPT-5.4 mini, GPT-5 Mini, GPT-4.1 Mini) — fast and cheap. Summaries, classification, extraction, first drafts, chat features in your own app.
  • Nano (GPT-5.4 Nano, GPT-5 Nano, GPT-4.1 Nano, GPT-5.6 Luna) — the cheapest hosted tier. High-volume, low-stakes jobs: tagging, routing, simple rewrites.
  • Pro (GPT-5.5 Pro, GPT-5.4 Pro) — the flagship allowed to think much longer. Priced far above everything else; reserve for problems you would otherwise pay a specialist to solve.
  • Open-weight (gpt-oss-120b, gpt-oss-20b) — models OpenAI released for anyone to host. Available through the API-compatible catalogue at very low prices, or run on your own hardware.
  • Reasoning line (o3, o4 Mini) — older dedicated reasoning models. Still listed; the GPT-5.x line has largely absorbed their role.
Prices from our catalogue at the time of writing, $ per 1M tokens. The price-ladder figure above is always the live source.
ModelInput $/1MOutput $/1MContext windowPick it for
GPT-5.55.0030.001,050,000Hardest reasoning, high-stakes output
GPT-5.42.5015.001,050,000Everyday professional work
GPT-5.6 Sol2.0010.001,050,000Standard-tier alternative
GPT-4.12.008.001,047,576Long-document work on a budget
GPT-5.4 mini0.754.50400,000Summaries, extraction, app features
GPT-5 Mini0.252.00400,000Cheap general tasks
GPT-5.4 Nano0.201.25400,000High-volume classification
GPT-5 Nano0.050.40400,000Cheapest hosted OpenAI text model
gpt-oss-120b0.0370.17131,072Open-weight, bulk jobs, self-hosting

Context windows: what the number means

The context window is the maximum amount of text (prompt plus reply) a model can hold at once, measured in tokens. A token is roughly three-quarters of an English word. A 400,000-token window fits a few novels; the 1,050,000-token window on GPT-5.4 and GPT-5.5 fits a small codebase or a year of meeting transcripts.

Bigger is not free. Every token in the window is billed as input on every turn, and models get slower and sometimes less accurate as the window fills. The practical rule: paste what the model needs, not everything you have. Lesson 7 covers the token maths in detail.

Inside ChatGPT you do not see per-token prices; you see a model picker and usage limits per plan. The lesson still applies. Picking a lighter model inside ChatGPT stretches your plan's limits, and picking a heavier one burns through them faster. Compare the full catalogue, including other providers, on /models.

Knowledge check

You need to tag 50,000 support tickets by topic. Which OpenAI tier is the sensible starting point?

Lesson FAQ

Which model does ChatGPT use by default?

It depends on your plan and OpenAI's current routing. Free users get a capable current model with limits; paid plans expose a picker with stronger models and, on Pro, the Pro-tier reasoning modes. The exact lineup changes, so check the picker in the app.

Is GPT-5.5 worth six times GPT-5.4 mini?

Only for tasks where GPT-5.4 mini fails. On simple writing and summarisation, the cheaper model is usually indistinguishable. Test both on your real task before paying flagship prices by default.

What is gpt-oss?

gpt-oss-120b and gpt-oss-20b are open-weight models OpenAI released publicly. You can run them yourself or use them through API-compatible hosts at very low per-token prices. See /tutorials/llama-101 for the open-weight approach in general.

Finished reading?

Mark it done to track your progress through the course.

Compare, calculate, decide — for ChatGPT

If our calculators helped you cut down on hidden AI wallet leaks, consider buying us a coffee. A tiny fraction of your savings keeps our pricing indexes updated daily.

Not sure which AI is cheapest for your use case? Find out in 30 seconds — no signup required.

AI cost intelligence

Stop overpaying for AI tools

Join the OverpayingForAI list for pricing updates, cheaper alternatives, and practical buying guidance.

Now tracking 50+ AI tools, models, platforms, subscriptions, coding tools, and automation products.

We use your email only for OverpayingForAI updates. Unsubscribe anytime.