MeteGPT API

Call the rewriting engine from wherever your text already lives. You hand it a passage, collect the result a little later, and pay for the words you sent. Checking a passage for AI never costs a credit.

Get a key See the rates

Authentication

Send a bearer key on every call. Keys are issued from the API tab in your account. You see the value once, when it is created; after that only its hash exists here, so there is nothing for us to hand back if you lose it. Keep up to five live at a time and you can retire one without a gap in service.

Authorization: Bearer mg_live_...

Jobs, not blocking calls

A rewrite is six passes over the text, not one, and lands somewhere between half a minute and four minutes depending on how much you sent. Nothing about that fits inside a normal HTTP request: our CDN drops origin connections near the 100-second mark, Zapier gives an action 30 seconds, Make allows 40 by default. So you get an id back straight away and read the answer from a second endpoint. Checking once every five seconds is plenty.

Send a passage

POST /api/v1/humanize

Twenty words is the floor and 3,000 the ceiling for one call. Leave mode out for deep, which is the six-pass run; fast does three and costs half as much. Nothing leaves your balance at this step.

curl -X POST https://metegpt.com/api/v1/humanize \
  -H "Authorization: Bearer mg_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "At least twenty words of the passage you would like rewritten.",
    "mode": "deep"
  }'

202 Accepted

{
  "id": "3d81f7ac-...",
  "status": "queued",
  "mode": "deep",
  "words_in": 640,
  "words_to_bill": 1000,
  "min_billable_words": 500,
  "poll_url": "/api/v1/jobs/3d81f7ac-..."
}

Read the answer

GET /api/v1/jobs/{id}

You will see processing, then either completed or failed. The charge happens on whichever poll first sees the job land, and only that one: ask again afterwards and you get the same text with nothing further taken. A job that broke, or whose output our gates refused, is never charged for.

curl https://metegpt.com/api/v1/jobs/3d81f7ac-... \
  -H "Authorization: Bearer mg_live_YOUR_KEY"

{
  "id": "3d81f7ac-...",
  "status": "completed",
  "mode": "deep",
  "output": "The rewritten passage.",
  "words_in": 640,
  "words_out": 828,
  "words_billed": 1000,
  "min_billable_words": 500,
  "words_remaining": 199000,
  "quality": "verified"
}

quality comes back as verified when every chunk passed the engine’s own checks on meaning, figures and length, and best_effort when one or more did not. You get the rewrite either way; the field is there so you know which ones to read before publishing.

Check text for AI

POST /api/v1/detect

No charge, ever, and it does not draw on your words. Under 60 words the model declines to answer instead of guessing from too small a sample, so short passages come back as a 400 rather than a confident-looking number.

curl -X POST https://metegpt.com/api/v1/detect \
  -H "Authorization: Bearer mg_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "text": "Sixty words or more, please." }'

{
  "probability": 0.11,
  "verdict": "Human",
  "verdict_code": "HUMAN-LIKELY",
  "words": 190,
  "billed": false
}

What it costs

A credit is a word. There is no conversion rate to work out, because a hidden multiplier is only ever a price you cannot compare against anyone else’s.

Calls round up to the next 500 words. Send 200 and you are charged 500; send 640 and you are charged 1,000. Before a single word is touched, each call spends a fixed stretch of GPU time, which is why 250 words at a time genuinely costs several times more per word than 3,000 does. Rounding puts that in the open rather than folding it into the headline rate.
The words you send set the price. Rewrites usually come back longer than what went in, and that growth is ours to carry: the words_to_bill figure you get at submission is what you pay, whatever length comes back.
Fast mode is half price. Three passes in place of six.
Only successes are charged. Breakages, timeouts and rejected output are free.
The allowance restarts, it does not accumulate. Whatever you have not used when the billing date comes round is gone, and that is what the per-word rate is built on. Take the tier that matches your real volume.

Plans

One list, charged monthly, identical to the rates on our sister site humanizemy.ai because both run the same engine. Outgrow a tier in the middle of a month and you move up: the bigger allowance is yours immediately and the difference appears on your next invoice.

Plan	Words / month	Price	Per 1,000
200k words	200,000	$59/mo	$0.295
400k words	400,000	$99/mo	$0.248
600k words	600,000	$135/mo	$0.225
800k words	800,000	$165/mo	$0.206
1,000k words	1,000,000	$189/mo	$0.189

Beyond a million words a month we set the rate in a contract, and it goes down to $0.138 per 1,000. For that, or for anything this page does not answer, write to hello@metegpt.com.

Errors

Failures arrive as { error: { type, message } }. Branch on type rather than on the wording, which we may improve.

Code	Type	When
400	invalid_request	Body will not parse, fewer than 20 words, or beyond the 3,000-word limit for one call.
401	authentication_error	No key, an unrecognised key, or one that has been revoked.
402	insufficient_credits	Not enough words left for this call. words_required and words_remaining come back in the body.
404	not_found	That job id does not belong to the account the key belongs to.
503	service_unavailable	Jobs are not being accepted at the moment. Back off and try again.

Things we will not claim

Your passage is dropped once the job has been paid for or has failed.
Nothing sent through this API is used to train anything.
There is no latency guarantee here, and we are not going to invent one. GPUs scale down to nothing between jobs, so the first call after a quiet stretch is slower than the ones behind it.
Reselling this as a rival rewriting service is not allowed. Putting it inside your own product is precisely the intended use; get in touch and we will agree terms.

Larger volumes, or anything the reference leaves open: hello@metegpt.com.