Claude vs ChatGPT in 2026, plans and real costs

Both standard plans cost $20 a month, and both will answer most questions well. The differences are in what that $20 buys, which models you actually get, and how the limits work. Here is how they compare as of 3 October 2026, from Anthropic’s and OpenAI’s own pages.

We don’t rank them on quality. Every benchmark you’ll find tests someone else’s prompts. The last section shows how to put your own question to both and compare the answers and the bill.

The short version

• Pick Claude if you work with long documents or code. Its top three models get a 1M-token context window in paid chats, and Claude Code comes with every paid plan.
• Pick ChatGPT if you want images and voice in the same app, or a cheaper way in. It generates images, has an $8 Go plan in the US, and the free tier offers unlimited text chats.
• Pick neither if you’d rather pay per answer. Both families are on the API at the same price for their mid-tier models.

Plans and prices

Claude’s yearly discount is the one structural price gap. ChatGPT Plus and Pro are monthly only, and so is Claude Max.

Which models you actually get

Claude. Free gets Sonnet and Haiku, but not Opus. Pro and Max add Claude Opus 5.5. Claude Fable 5.1, Anthropic’s top-priced model, is on every paid plan, though Pro pays for it through usage credits. Max can spend up to half its weekly limits on Fable at no extra cost.

ChatGPT. Free and Go run GPT-5.6 Luna. Paid plans run GPT-5.6 Sol in regular chats, at several reasoning levels (Plus gets Medium and High). GPT-6 Pro, built on GPT-6 Astra, is on Pro $100, Pro $200, Business and Enterprise. One thing surprises people. The GPT-6 Sol and Luna models aren’t in regular ChatGPT conversations at all. They live in ChatGPT Work and Codex.

Context. Claude Fable 5.1, Opus 5.5 and Sonnet 5.5 each get a 1M-token window in paid chats. ChatGPT’s reasoning modes get 256K on Plus and 400K on Pro, and its instant mode less than that. On the API, both families reach about a million tokens.

Limits

Claude’s limits reset on a rolling five-hour window, and paid plans add weekly limits on top. Pro gets at least five times Free’s usage per session, and Max gets five or twenty times Pro’s. Anthropic raised the five-hour limits when Opus 5.5 launched in September but didn’t say by how much.

OpenAI publishes no message counts. Its plans say Limited, Expanded or Maximum, and Plus “may include usage limits such as message caps”, especially at busy times.

Features that differ

• Images. ChatGPT generates them. Claude doesn’t make photos or illustrations, though it builds charts and diagrams as code.
• Voice. Both have it. ChatGPT sets daily hours, three on Plus. Claude’s voice mode is in beta on every plan.
• In the browser. Claude in Chrome comes with every paid Claude plan. OpenAI’s extension runs in Chrome, Edge, Brave, Opera and Vivaldi. We wrote guides to Claude in Chrome and the ChatGPT extensions.
• Coding. Claude Code is in every paid Claude plan. ChatGPT has Codex, limited on Free and Go and expanded on Plus.
• Ads. ChatGPT’s Free and Go plans may show ads. Its paid plans don’t.
• Your data. Both train on consumer chats unless you switch it off. In Claude it’s the model training setting under Privacy. In ChatGPT it’s “Improve the model for everyone”. Team and Business accounts aren’t trained on by default.

What the same answer costs on the API

Skip the subscription and you pay by the token. These are list prices per million tokens on 3 October 2026, the same on OpenRouter as on Anthropic’s and OpenAI’s own pricing pages.

The last column is our estimate for one question about a web page. That’s about 5,500 tokens in, counting the page text and the app’s own instructions, and 250 tokens out. At list price, $20, the cost of either subscription, covers about 1,500 of those questions on Sonnet 5.5 or GPT-6.1 Sol, and about 300 on Fable or Astra.

Real bills run higher. A follow-up resends the whole conversation, reasoning tokens bill as output, and faster routes can cost more than list price. Light users usually come out ahead on the API. Someone who chats all day with a frontier model often doesn’t.

Ask both in your browser

The fair test is your own question, on the page you’re reading, sent to both. That’s what UtilEngine, our extension for Chrome, Brave and Edge, is built for. One OpenRouter key reaches 85 models, including 15 from Anthropic and 36 from OpenAI. Ask with Claude Sonnet 5.5, switch the chat to GPT-6.1 Sol, and ask again.

Each answer bills to your OpenRouter account. Add credits before you paste the key, since UtilEngine checks it with one small paid request. Most models run on OpenRouter’s faster route, which can cost more than list price. UtilEngine itself is $49 once. The bring-your-own-key page explains the split.

Common questions

Is Claude better than ChatGPT?

For some tasks, and the reverse for others. Claude’s paid chats hold far more text at once, and ChatGPT makes images. For writing and reasoning, run your own prompts through both.

Which is cheaper?

At the standard tier they cost the same, $20 a month. ChatGPT has the cheaper ways in, with unlimited free text on GPT-5.6 Luna and an $8 Go plan, both with ads. On the API, Sonnet 5.5 and GPT-6.1 Sol share a list price.

Can I use Claude and ChatGPT together?

Yes. A router like OpenRouter gives you both families under one key and one balance, and any app that takes an OpenRouter key can switch between them. Our guide on how to use OpenRouter covers the setup.

Do Claude and ChatGPT train on my chats?

By default, on consumer plans, yes for both, with an opt-out in settings. Claude’s Incognito chats aren’t used for training. Business and team plans on both sides are excluded by default.

Which has the bigger context window?

In the chat apps, Claude. Its top models get 1M tokens on paid plans, against 256K to 400K in ChatGPT’s reasoning modes. On the API they’re close, at about a million tokens each.