AI that can analyze images for free: ChatGPT allows 3 a day, Claude and Copilot take 20 a chat

Which AI can analyze images for free, and how many?

Size caps swing even wider than the counts, from 20 MB per image on ChatGPT to 500 MB per file on Claude. Each row below comes from the vendor's own help page, read on 12 October 2026.

ChatGPT is the tight one. OpenAI's file uploads FAQ says "Free users are limited to 3 file uploads per day," and a photo is a file upload like any other. The general cap is 80 files every 3 hours, and OpenAI says it may lower these limits at peak hours.

Claude and Copilot are the roomiest. Anthropic's upload article allows 20 files per chat and images up to 8,000 by 8,000 pixels. Microsoft's Copilot upload page also allows 20 files a conversation, at up to 50 MB each, and says it covers free Copilot Chat. Gemini takes 10 files in one prompt and up to 100 MB a file. Google's Gemini upload page adds "rolling limits" on how much you can analyze, and it doesn't give a number.

How do you get an AI to analyze an image and answer questions?

You attach the image, then ask one specific question about it. A bare "what is this" gets a bare caption back.

1. Open ChatGPT, Gemini, Claude or Copilot and click the + or paperclip in the message box. Claude also lets you drag a file in or paste a screenshot from your clipboard.
2. Crop to the part you're asking about. Anthropic recommends images of at least 1,000 by 1,000 pixels, and OpenAI tells you to enlarge small text in the image.
3. Give the question a job: what to read, what to compare, or what to check.
4. Ask the model to list what it sees before it concludes anything.
5. Check every count, total and name against the image yourself.

These five prompts cover most of what people upload. Fill the brackets.

Read the text: "Transcribe every word in this image exactly as written, line by line. Mark anything you can't read as [unclear] instead of guessing."

Read a chart: "Turn this chart into a table with one row per data point. Give the axis labels and units. Say which values are printed on the chart and which you estimated from the gridlines."

Fix an error: "This is a screenshot of an error. Quote the error text, tell me which app shows it, and give me the three likeliest fixes in order."

Spot the difference: "These two photos show the same [room, product, document]. List every difference you can see and where it is in each image."

Identify something: "What is this [plant, part, dish]? Give your best guess, two alternatives, and the visible features that tell them apart."

The last sentence of each prompt does the work. Asking for visible evidence first stops the model jumping to an answer, and our multimodal AI guide builds a whole workflow around that habit.

What can't AI see in an image?

AI can't read a photo's metadata, and it guesses at counts. OpenAI's vision guide lists 11 limitations as of October 2026. Most of them apply to any chatbot you'd try:

• It doesn't process file names or metadata, so the date and GPS location your camera saved stay unread.
• It "may give approximate counts" of objects in a photo.
• It may misread rotated or upside down text and images.
• It can stumble on text in scripts such as Japanese or Korean, and on graphs whose lines differ only by color or dash style.
• It struggles with panoramic and fisheye shots, and with exact positions, like the pieces on a chess board.
• OpenAI blocks CAPTCHA images outright.

Medical scans are the hard line. OpenAI says its model is "not suitable for interpreting specialized medical images like CT scans" and shouldn't be used for medical advice. An X-ray or a photo of a rash belongs with a clinician.

Long PDFs lose their pictures, too. Anthropic says Claude reads the charts and images in a PDF of 100 pages or fewer. From 101 to 1,000 pages, it reads the text only.

Which AI is best at analyzing images?

Claude Opus is the best image reader on the leading public benchmark, and Copilot or Claude is the best free pick for volume. On Artificial Analysis's MMMU-Pro leaderboard, read on 12 October 2026, Claude Opus 5.5 scores 88% and OpenAI's GPT-6 Astra 87%. Both sit behind paid plans.

MMMU-Pro is a hard test to game. It has 3,460 questions across 30 academic subjects, and it drops any question a text-only model can answer. Some questions arrive only as a screenshot or photo, so a model has to read the image to score.

For free use, the rule is simple. Use Copilot or Claude for a batch of up to 20 images in one chat. Use Gemini for big files, up to 100 MB each. Keep ChatGPT Free for the odd photo, since 3 uploads a day runs out by lunch. If you want to change the picture rather than read it, editing photos with AI is a different set of tools.

Whizi, which publishes this page, sends an attached photo to whichever model you pick, so on Pro you can ask Claude Opus and Gemini about the same image side by side. It loses on the free question. A free Whizi account gets 10 text messages a day on Whizi AI, Auto or Claude Haiku 5.5, with no file uploads, so a batch of photos needs a plan. The website takes up to 6 files a message but caps each image at 8 MB and 60 image uploads a day, refuses GIFs, and doesn't flag which of its 280+ models can see images. Supported file types lists the rest.

Can you analyze images with an API instead of a chatbot?

Yes, and the big cloud vision APIs include a free monthly allowance. Google Cloud Vision gives the first 1,000 units a month free, then charges $1.50 per 1,000 for label detection, per its pricing page. Microsoft's Azure AI Vision free tier allows 5,000 transactions a month at 20 a minute. Amazon Rekognition is the third big option.

These services return labels, text and object boxes as data, not chat answers. That suits an app that tags thousands of product photos. It's overkill for one screenshot.

DeepSeek came late. It opened image input on its API on 21 August 2026, and its vision guide accepts up to 600 images in one request on its Flash model. Its pricing page still marks vision "Not supported" on its flagship V4 Pro, and DeepSeek's docs say nothing about images in its own chat app.