Improved
Parsing API now reads text up to 8,192 characters with Atlas
26 days ago by ReadMe API
POST /text/parsing now reads each text up to 8,192 characters with Atlas and up to 20,000 with Quicksilver, so long job postings and resumes are read to their end. Longer texts are still cut at these lengths.
Action required
- Now: none.
POST /text/parsingkeeps its parameters, their defaults and its response. - Later: none. Nothing in this update is deprecated.
What changed
- Atlas (
parsing_model=atlas): each text is read up to 8,192 characters, up from 4,000. Most long job postings are now read to their end: their copied sections stop mid-sentence far less often, and the sections written late in a posting, such as the benefits and the interview process, are filled more often. - Quicksilver (
parsing_model=quicksilver): each text is read up to 20,000 characters, up from 4,000, whatever the number of texts in the request, so entities written late in a long resume or job posting are returned too. - Response time: with Atlas, a text longer than 4,000 characters takes a little longer to answer, since more of it is read. Shorter texts answer as before.
Performance
| Atlas, time to read a job posting's sections | Before | Now | Change |
|---|---|---|---|
| Median | 1.49 s | 1.5 s | no change |
| 99th percentile | 3.03 s | 4.11 s | 36% more time |
Measured on our servers for every job posting parsed with Atlas, on the five weekdays before the change (21 to 27 October 2026) and the six weekdays after it (28 October to 4 November 2026). "No change" means the new figure lies within the day-to-day spread of the old one.
Example
curl -X POST https://api.hrflow.ai/v1/text/parsing \
-H "X-API-KEY: <your_api_key>" \
-H "X-USER-EMAIL: <your_email>" \
-H "Content-Type: application/json" \
-d '{
"texts": ["<a job posting of up to 8,192 characters>"],
"parsing_model": "atlas",
"output_object": "job"
}'The response keeps the format documented in the reference: one parsed job per text, with sections that now run to the end of a long posting.