AmberPen Get started
NEURAL PROOFREADING API

The proofreading API for flawless writing.

AmberPen is a neural grammar checker that returns spelling, grammar, and style corrections as structured edits in real time.

Get started
Proofreading mode
Fast, single-pass correction
Your text
141 / 2,000 characters
JSON output 
The API response will appear here.
Low latency
JSON output
Structured edits
BUILT FOR EVERY WRITING EXPERIENCE

Grammar checking for every product that handles text.

</>

For writing apps

Add real-time proofreading to your editor, inbox, or writing assistant.
Get started

For education tools

Help students write with confidence and clarity.
Get started

For CMSs

Deliver clean, consistent content across your platform.
Get started
BENCHMARKS

Measured against the field.

We ran AmberPen, Sapling, LanguageTool, and GrammarBot over the complete JFLEG and CWEB public benchmarks. Pick a metric to compare.

✓ Best JFLEG fluency score
✓ Highest recall on CWEB web text
✓ Fastest neural API at p50
✓ Only API with explanations + style preservation
Complete runs, July 2026. Methodology and full results
HIGHER IS BETTER
AmberPen
0.561
Sapling
0.470
GrammarBot
0.413
LanguageTool
0.307
Official GLEU on all 754 JFLEG dev sentences.
JFLEG rewards grammar correction and broader fluency improvements.
AMBERPEN LEADS
EVERYTHING AN API SHOULD BE

Built for the details other proofreading APIs skip.

Streaming corrections

Edits arrive as they are found, so even a book-length text gives feedback immediately.

Structured edits

Every correction returned as a JSON diff with offsets—render it your way.

Custom dictionaries

Per-product and per-user terms, so brand names never get “fixed”.

Explanations built in

Evaluate mode reviews every proposed edit and adds a plain-language reason.

Private by default

No training on your data. Nothing retained once the response is sent.

Predictable cost

Flat per-character pricing. No surprise bills, no token math.
ONLY ON AMBERPEN

Incremental & live correction

Resend a large document and AmberPen proofreads only what changed—the untouched remainder never goes back through a model. That means true as-you-type live correction, and thousands of dollars saved on long documents.
DOCUMENT RESUBMITTED · 48,000 CHARS
Changed — proofread
Unchanged — skipped
3% reprocessed · 97% saved
~250ms
PREDICTABLE PRICING

Flat per-character. No token math.

Starter

€5 /mo
500K characters included, then €9 per million
✓ Full corrections API
✓ Explanations included
✓ Community support
Get started
MOST POPULAR

Pro

€49 /mo
9M characters included, then €6 per million
✓ Custom dictionaries
✓ Streaming mode
✓ Incremental proofreading
✓ Email support
Get started

Scale

Custom
Volume pricing from 50M characters
✓ Dedicated latency SLA
✓ SSO & audit logs
✓ Private deployment options
Talk to us

Elevate your writing.

STATE-OF-THE-ART CORRECTION

Corrections a rule-based checker can’t reach.

Real outputs from the AmberPen API, unedited. Every one of these needs an understanding of the whole text, not a dictionary and a pattern list.

Seamless language mixing

BEFORE

Our Berlin team wrote: Ich habe das Dokument gestern gesendet an dich, so we need to recieve a new quote.

AFTER

Our Berlin team wrote: Ich habe das Dokument gestern an dich gesendet, so we need to receive a new quote.

German word order and an English typo, fixed in the same pass.

Context-aware corrections

BEFORE

Conversion rose from 2% to 4, which is a solid gain.

AFTER

Conversion rose from 2% to 4%, which is a solid gain.

Nothing here is misspelled. The missing unit only exists in the rest of the sentence.

Reasoning across sentences

BEFORE

Anna forgot his laptop at the office. He said she would pick it up tomorow.

AFTER

Anna forgot her laptop at the office. She said she would pick it up tomorrow.

Both pronouns are ordinary English. Only the antecedent in the previous sentence makes them wrong.

Real words in the wrong place

BEFORE

Neither the developers nor the designer have finished there part of the sprint.

AFTER

Neither the developers nor the designer has finished their part of the sprint.

Every word passes a dictionary check. Agreement here follows a subject four words back.

Ship better writing experiences.

Try for free with a test api key. Integrate in minutes.

Start now