AI Chatbots Compared: Limits, Price, File Upload and Web Search

Five-hour windows with a weekly cap are now the norm

Claude paid plans use a session limit that resets five hours after your first message, and a weekly limit that resets at a fixed time assigned to your account. Gemini works the same way: Google says the limit refreshes every 5 hours until you reach the weekly limit. ChatGPT Plus is described by OpenAI as message estimates per 5 hours, with weekly limits that may also apply.

The practical meaning is simple. Starting a heavy session at 9:00 means the window closes at 14:00, so a big task is better begun when you can use the whole window.

Two plans break the pattern. Claude Ultra, at $200 a month, has no 5-hour limit for verified accounts, and OpenAI says its Pro plans currently have none either.

Multipliers: how plans scale without publishing a number

Anthropic sells Claude Pro as the 1x baseline, Max at $100 a month as 5x, and Ultra at $200 as 10x per week. Google rates its plans against a standard limit: AI Plus is 2x, AI Pro is 4x, and AI Ultra is 5x or 20x of AI Pro depending on the tier. Mistral says Pro gives up to 6x the messages, up to 5x the web searches and up to 40x the image generation of the free plan.

None of these pages convert the multiplier into a prompt count. Google says limits depend on request complexity, the features used and the length of the conversation, and may change without notice.

ChatGPT counts messages per model, and the range is wide

OpenAI's docs publish estimates for Plus that depend on the model: GPT-6 Astra 5 to 45 messages per 5 hours, GPT-6.1 Sol 15 to 160, GPT-6.1 Terra 35 to 285 and GPT-6 Luna 350 to 3,000. One plan spans 5 to 3,000 messages depending on the model, so the model you pick matters more than the plan.

These figures come from OpenAI's Codex and ChatGPT docs page. OpenAI's main pricing page and help center returned an access error on 2026-11-01, so file, voice and search limits for ChatGPT are not covered here.

Context window: 32k, 128k, 200K, 500K or 1M tokens

A larger window lets one chat hold more text before older parts drop out. Gemini lists 32k tokens without a plan, 128k on AI Plus and 1 million on AI Pro and Ultra. Claude lists up to 1M tokens for the newest models on paid plans and 500K or 200K tokens for other models.

Anthropic adds that part of the window is always reserved for the reply, so the longest usable conversation is a little shorter than the headline figure.

File upload rules differ more than prices do

Claude documents its caps in numbers. A chat takes up to 20 files at 500MB each, images up to 8000x8000 pixels, and PDFs up to 1000 pages. Project files are limited to 30MB each. Supported documents include PDF, DOCX, CSV, TXT, HTML, ODT, RTF, EPUB, JSON and XLSX, and images can be JPEG, PNG, GIF or WebP.

Mistral counts storage rather than files: document libraries reach 15GB on Pro and 30GB per user on Team. Gemini's limits page gives no file count or size, only a note that a 100-page PDF uses more of your allowance than a short one.

A worked example: one 400-page PDF through three assistants

Take a 400-page report, 40MB as a file. On Claude, it is under the 500MB cap and under the 1000-page cap, but pages 101 and above are read as text only, so charts after page 100 are not analysed. To keep charts, split it into four 100-page files, which stays inside the 20-file chat limit.

On Gemini, the page states no size limit, so the deciding number is the context window: 32k tokens without a plan is small for 400 pages, and AI Pro offers 1 million. With Mistral Pro, the 40MB file fits easily inside a 15GB library.

Web search and citations: what each page actually promises

Claude Free lists web search. Mistral lists web searches on Free and up to 5x more on Pro. Perplexity says its Best mode has no quota limit and gives 10x as many citations per answer, but it does not print a price on the help page we read. Microsoft's Copilot page lists a free tier with a Microsoft account and no figures.

Only Perplexity puts citations in its pitch, so for work where you must check sources, open the linked pages rather than trusting the summary.

Entry prices, as listed on 2026-11-01

ChatGPT Go is $8 a month and Plus $20. Claude Pro is $20 monthly or $17 on an annual plan. Mistral Pro is $14.99 before taxes. Google's Estonian page lists AI Plus at EUR 4.99 and AI Pro at EUR 21.99, both in euros.

Perplexity and Copilot prices were not on the pages we could read, so they are left out rather than guessed.

Mistakes when reading a comparison table

Treating 4x as a promise of four times more work is the first one, because the base is unstated. Mixing currencies is the second: EUR 21.99 is not $21.99. Third, ignoring the model: ChatGPT's estimate for one plan runs from 5 to 3,000 depending on it.

If a plan page confuses you, Crade can read it from your screen and explain what each row means.

Which kind of user each limit model suits

Short daily questions suit any free tier, since the 5-hour window rarely binds. Long documents suit the plans with a stated file cap and a large window. Heavy all-day work suits the tiers with no 5-hour limit, at $200 a month and up.