ChatGPT

What it actually does

ChatGPT is a general-purpose assistant you talk to in a browser or an app. On the paid tier it runs GPT-4o with vision, so you can upload an image and ask questions about it, plus voice mode, a Code Interpreter that runs Python against files you upload, and DALL-E 3 image generation folded into the chat.

Custom GPTs let you save a prompt and a set of files as a reusable assistant, and the GPT Store offers pre-built ones. That is the whole product. Most of the other tools on this site are either a wrapper around a model like this one or a narrower product competing with something ChatGPT already ships.

What it costs

The free tier gives you GPT-4o mini with a message limit. Plus is $20 a month. Team is $25 per seat per month. Enterprise is quote-only.

The catch is that $20 does not buy a fixed entitlement. Rate limits move. GPT-5 Thinking launched capped at 200 messages a week and only went to 3,000 after users complained, and OpenAI's own wording was that limits "may need further adjustments depending on usage". The older 4o model is capped at 80 messages every three hours on Plus, and that cap has been cut before, per OpenAI's community forum threads.

Model choice is also a paywall. GPT-4.5 was held back for the $200 a month Pro plan on GPU cost grounds, and the "Show additional models" toggle is Pro-only. If one specific model matters to your work, Plus does not guarantee you keep it.

Where it falls down

The August 7 2025 GPT-5 launch is the thing a buyer should read about first. OpenAI removed every previous model from ChatGPT on the day, pushing Plus subscribers onto the new one without warning. Ars Technica reported it as one of the most vocal user revolts in the product's history, and a "GPT-5 is horrible" Reddit thread passed 2,000 comments in days.

It got worse before it got better. The automatic router that decided which GPT-5 variant answered you malfunctioned and kept falling back to less capable variants, so paying subscribers got worse answers than they had the week before. OpenAI reversed within a week and restored 4o for paid users.

That is a pattern rather than a one-off. o3 was retired on August 26 2026. The DALL-E GPT shut down on August 30 2026. ChatGPT Atlas went on August 9 2026. Sora 2 is scheduled for full shutdown on September 24 2026. On the API side, the chatgpt-4o endpoint deprecation dated February 17 2026 drew complaints that the replacement refused more requests.

Hallucination and the knowledge cutoff are still real limits, and the free tier's message caps make it a demo rather than a working tool.

Who should buy it

Almost everyone, which is why this scores 4.0 and not lower. G2 has it at 4.6 out of 5 from 2,996 reviews with 78% five-star, and that matches what the product delivers: one $20 subscription that replaces several $20 subscriptions elsewhere.

Do not buy it if you need a stable, reproducible model over a period of years. Anything carrying a compliance or audit requirement, or a workflow built on one model behaving a particular way, belongs on the API with a pinned model version and a written migration plan, and ideally with a second provider you can switch to. OpenAI does respond to loud feedback (4o came back, Thinking limits went up 15 times, routing controls were added), but the response arrives after the removal, never before it.

If you are on the free tier and hitting the message limit most weeks, pay the $20. If you are eyeing the $200 Pro plan purely to get model choice back, price out the API first, because that is the same problem with a different bill attached.