AI Text Summarizer
Summarize a long article, an email thread, or a report down to a few sentences. Pick short, medium, or long; the AI model downloads once and then works offline.
Summary
The summary lands here.
The summary stopped
Canceled.
How to use
- Paste the article, thread, or report into Text to summarize.
- Pick a Summary length, then look at the Model cards. Sharp is chosen wherever it can run; where it cannot, its card says why and Standard stays chosen.
- Press Summarize. Compare what lands in the Summary pane against the original, then press Copy summary.
About this tool
Some documents you only need the gist of. A forty-paragraph release note. A mail thread you got added to at five on a Friday. A paper you have to cite but will never read closely. Paste it into the text summarizer, choose a length, press Summarize. Short comes back as a sentence or two, medium as three or four, long as six to eight. You watch the summary get written a few words at a time, and there is a Cancel button beside the progress bar if it starts wandering. Under the summary, the report says how far it shrank your text.
Two models sit behind the Model cards, and each card prints what it costs to fetch and whether it is already on this device. Standard is a distilled BART, an older summarizing model that loads on any browser including Safari and phones. Sharp is Qwen3-0.6B, a small language model that reads the passage and then writes about it, which is why it holds together over longer text. Sharp wants WebGPU, so the page chooses it for you when your hardware can reach the graphics card and keeps Standard when it cannot. Whichever one you use downloads a single time, then stays on the device and works with the network off.
A 2,000-word release note, run through both models
Paste a release note of about 10,000 characters, close to 2,000 words, and leave the length on Medium. With Standard chosen the page cuts the text into four pieces of up to 3,600 characters, each ending at a sentence end and overlapping the next by whole sentences, and asks the model for 33 to 70 tokens on each. Each piece streams into its own row. The four partial summaries are then joined and summarized once more inside the medium window of 55 to 100 tokens, five passes in all, and the result is trimmed to its first four complete sentences; anything the trim drops shows as a Cut row that Copy summary leaves out.
Switch to Sharp and the shape changes. Sharp reads the first 1,000 words, so the page marks the word where it stops, names the two sections it never reaches, and offers Summarize by section, which reads each heading's section whole in its own pass and tags each summary with the heading. Either way the model is asked for three or four sentences under a 224-token ceiling, and the summary appears a few words at a time while it writes. The first run of a model begins with Fetching the model and a share of its size; once the files are on the device it says Opening the model, then Summarizing. Press Cancel partway and the page says Canceled, keeping whatever had streamed.
Good to know
- Sharp reads the first 1,000 words of your paste and nothing after them. The page marks the word where it stops and names the sections it never reaches; when the text has headings, Summarize by section reads each section whole in its own pass.
- Standard's long-text route squeezes each piece of up to 3,600 characters down to 50, 70 or 126 tokens, depending on the length you chose, before it has seen the whole. A fact that appears once, deep in a long paste, can be lost at that first stage and never reach the final summary, and a paste long enough to need a second round of shortening loses a little more.
- Standard is trimmed by sentence count after the fact: two sentences for short, four for medium, everything it wrote for long. The count uses the browser's sentence segmenter, so “Dr.” and “e.g.” stay inside their sentences, and a final sentence the model left dangling is dropped rather than shown half-built. Whatever the trim drops shows as a Cut row, and Copy summary leaves it out.
- Standard is a 2019-era summarizer. On dense or technical prose it will flatten a qualifier, or state something the source never quite said.
- Sharp is 0.6 billion parameters. It stays much closer to the source than Standard does, and it is still small enough to misread a contract clause or a methods section.
- That first download hurts on a metered connection, and Sharp, the better of the two, is about twice the size of Standard. Each card says what its model costs before you press.
- Output is one block of prose. There is no bullet mode, and no way to trace a claim back to the paragraph it came from.
Things people ask
Should I pick Standard or Sharp?
Take Sharp any time its card lets you. It reads the whole passage and writes about it, so it stays coherent on long input where Standard starts repeating the opening paragraph. Standard is there for machines and browsers with no usable GPU, which mostly means older phones and older Safari.
How much text can I paste at once?
The two models answer differently. Sharp stops reading at word 1,000, and the page marks that word in your text; when the text has headings, Summarize by section reads each section whole and tags each summary with its heading. Standard has no cap. Anything past 3,600 characters is split into pieces that end at sentence ends and overlap by whole sentences, each summarized on its own, and the partials are summarized again, first in groups if together they would overflow the model's 1,024-token window. A 5,000-word report goes through whole, and the report under the summary counts the passes.
Why is the model download so large?
The model is the download. Sharp's files come to about 550 MB and Standard's to about 275 MB, and the first run of either also brings the ONNX runtime, about 26 MB, which other tools on the site share. Each model card prints its total before you commit, and the badge above the cards counts the one you chose while it arrives. It happens once per browser. After that the files stay on the device and a summary starts in seconds.
The summary contains something the article never said. What happened?
Compact summarizers work by predicting likely wording, and a prediction can land on a claim that was never in the source. Standard does it more often than Sharp, especially on technical passages. Read the summary next to the original before you forward or cite it.
What do short, medium, and long actually change?
Both the instruction and the budget. Sharp is asked for one or two sentences, three or four, or six to eight, with a token ceiling of 96, 224, or 448 to match. Standard is given a token window instead, 15 to 45 for short, 55 to 100 for medium, 110 to 180 for long, and then a sentence trim. Its floor never asks for more than a fifth of the text it reads, so a short passage gets a short summary rather than padding. Real length still moves with the input, because the model picks its own stopping points.
Will the same text give the same summary twice?
Both models decode greedily with sampling switched off, so identical text with the same length and the same model produces the same summary on every run. Change a word and the wording may shift further than you expect. Standard also refuses to repeat any run of three tokens, which is why its output never loops a phrase and occasionally why it bends a proper name that legitimately needs the same three tokens twice.
Why is Sharp greyed out on my machine?
Sharp's card says cannot run here and gives the reason under its name. Needs a browser with WebGPU support means the browser cannot hand the model to your graphics card at all. Needs a GPU with 16-bit shader support means WebGPU is there but the adapter lacks the half-precision shaders the Sharp weights are packed for. In either case Standard stays selected and does the job with the older model.
Missing something in AI Text Summarizer? Suggest a feature →
Tell someone who needs this.