AI at work

ChatGPT Plus vs Claude Pro vs Google AI Pro: what $20 a month buys

ChatGPT Plus, Claude Pro and Google AI Pro all cost about $20 a month. What each plan's own page says about usage limits, context and features, side by side.

ChatGPT Plus vs Claude Pro vs Google AI Pro: what $20 a month buys

ChatGPT Plus and Claude Pro cost $20 a month and Google AI Pro costs $19.99, so price does not separate them. What does is how each vendor describes its limits: OpenAI publishes context window sizes per plan, Anthropic and Google describe usage as a multiple of their free plan, and none of the three publishes a fixed number of messages. This comparison uses only what each vendor’s pricing page said on October 26, 2026.

The three plans side by side

ChatGPT Plus Claude Pro Google AI Pro
Price (US, monthly) $20 $20, or $17 a month billed yearly ($200) $19.99
How usage is described “Expanded messages and uploads”; “Limits apply” “At least 5x more usage per 5-hour session than Free” “4x higher usage access than Free”
Limit window Not stated as a time window Per 5-hour session Free-plan limits refresh every 5 hours until a weekly limit is reached
Cheaper tier below it Go ($5.50 where we checked; priced by country) Free only Google AI Plus, $4.99
Next tier up Pro, from $100 Max 5x, $100 AI Ultra, from $99.99

Context: how much a chat can hold

OpenAI is the only one of the three that lists context windows per plan on its pricing page. For ChatGPT Plus the total context window is 54K tokens with GPT Instant and 256K with GPT Reasoning, which OpenAI translates to about 40 and about 320 pages of input. The same page notes that part of the window is used by system instructions, memory and the model’s reasoning, so the space for your own text is smaller than the total.

ChatGPT plan GPT Instant window GPT Reasoning window
Free 27K (about 12 pages of input) Varies
Go 54K (about 40 pages) 256K (about 320 pages)
Plus 54K (about 40 pages) 256K (about 320 pages)
Pro 128K (about 250 pages) 400K (about 680 pages)

Anthropic’s and Google’s plan pages do not state a context size per plan. On their APIs, Claude Sonnet 5.5 and Opus 5.5 have 1M-token windows and Gemini 4 Argon has 1,048,576, but an app plan does not have to give you the full API window, so treat the API figure as a ceiling, not a promise.

What each plan adds beyond chat

ChatGPT Plus

OpenAI lists advanced reasoning with GPT-6, more complex image creation, expanded deep research, memory and context, projects, scheduled tasks and custom GPTs, expanded Codex usage for coding, and ChatGPT Work on desktop, web and mobile.

Claude Pro

Anthropic frames Pro by usage: at least five times the Free plan’s usage in each 5-hour session. The Max plans multiply the Pro allowance by 5 or 20.

Google AI Pro

Google frames AI Pro as four times the Free plan’s usage in the Gemini app, with Ultra raising limits by 5 or 20 times over AI Pro.

Which one should you pick?

  • You need long documents in one chat: ChatGPT is the only one of the three that tells you the plan’s window in advance. If you work past 256K tokens often, Plus will not be enough.
  • You hit limits in bursts: Claude’s limits reset per 5-hour session, so heavy mornings and quiet afternoons fit it well; a weekly cap also applies on the Max plans.
  • You already pay for Google storage or use Gemini in Workspace: AI Pro is the natural choice, and AI Plus at $4.99 is the cheapest paid step of the three vendors.
  • You use AI lightly: paying per use on the API may cost less than $20. At 30 messages a day of medium length, Claude Sonnet 5.5 on the API comes to about $11 a month by our estimate; the break-even is around 55 messages a day.

The subscription or API tool runs that comparison for your own usage against every plan’s price.

Sources

All pages read on October 26, 2026. Plans change often; the plan finder shows the date its prices were last checked.

Free toolIs a monthly AI plan cheaper than paying per use?Compare for my usage
DailyTech Editorial Team

We explain AI with numbers you can check. Every price, limit and benchmark links to its source and the date we read it. How we work · Report an error