CaptchaAI
The Grid That Asks For A Number, Not A Bus.
BLS-style CAPTCHAs are not image-recognition puzzles in the usual sense. They show a 3x3 grid and an instruction naming a number, and the answer is a cell selection rather than an object class. CaptchaAI treats it as its own supported type, published at over 99% success with average latency under a second.
A numeric instruction is a different problem from 'select all bicycles'
Classic image grids ask you to classify objects. A BLS-style grid names a number and asks for the matching cell, so the work is reading the instruction and locating the value rather than recognising a category. Solvers tuned only for object classes handle it poorly, which is why it is listed as its own type.
- The instruction carries the answer, not the imagery
- Cell position matters, so the response shape differs
- Object-classification tuning does not transfer
- Listed and measured separately for that reason
>99% success
CaptchaAI's published figure for BLS
< 1 second
Published average latency
One endpoint
Same API as every supported type
Proxy per task
Submit proxy + proxytype when needed
Working with the BLS type
The same submit-and-poll flow as the rest of the API.
Submit the task
POST to in.php on ocr.captchaai.com with your 32-character key, the method for the type and its parameters. Add json=1 for a structured response, or leave it off for the legacy plain-text shape.
Poll for the result
Send action=get&id=<taskId> to res.php roughly every five seconds until it stops returning CAPCHA_NOT_READY. With a published sub-second average for this type, the poll cadence is usually the dominant cost, not the solve.
Handle the documented errors
ERROR_UNSOLVABLE, ERROR_ZERO_BALANCE and CAPCHA_NOT_READY are documented codes with distinct meanings. Branch on them rather than treating every non-success the same way - one is retryable, one is a billing state, one is just 'not yet'.
Add a proxy when the site fingerprints IPs
Tasks accept proxy and proxytype parameters, so the solve can originate from the same network context as the rest of your session where the target site cares about that consistency.
What is published about this type
From CaptchaAI's own supported-challenges list.
Listed as a first-class type
BLS Captcha appears in the supported-challenges list in its own right, described as '3x3 numeric-instruction grid CAPTCHAs optimized for BLS-style tasks'.
Published success above 99%
The site publishes >99% success for this type, alongside Image CAPTCHA and Grid Image which carry the same published success band.
Published average latency under one second
Listed at <1s average, in the same band as Grid Image and slower only than plain Image CAPTCHA at <0.5s.
No separate pricing
Thread-based plans state no surcharges by type, so BLS runs on the same threads as everything else rather than on a special rate.
Numeric-instruction grids are a niche worth naming
They appear in appointment and application portals rather than general consumer signup flows, which is exactly why generic solvers handle them badly - most training pressure goes to the challenges everyone sees.
- Appointment and application portals, not consumer signup
- Instruction text carries the answer, not the imagery
- Listed and measured as its own type
- Same threads, same endpoint, no surcharge
Why this way
The practical case, in four lines.
Treated as its own challenge type
BLS grids appear in the supported list in their own right, with their own published success and latency figures.
Published sub-second average
Listed at under one second average latency, in the same band as grid image solving.
No special integration
Same in.php submit and res.php poll as every other type - no separate SDK or endpoint.
No premium pricing
Thread-based plans state no surcharges by type, so this runs on the plan you already have.
Common questions
What is a BLS-style CAPTCHA?
A 3x3 grid paired with an instruction that names a number, where the answer is the cell matching that value. CaptchaAI lists it as its own supported type, distinct from image and object-classification grids.
How is it different from a normal image grid?
An image grid asks you to classify what is depicted. A numeric-instruction grid asks you to read an instruction and locate a value, so the reasoning and the response shape both differ. That is why it is measured separately.
Does it cost more than other types?
No. The published pricing states no surcharges by CAPTCHA type - every supported type runs on the same threads under the same plan.
What if a task comes back unsolvable?
ERROR_UNSOLVABLE is a documented response. Treat it as a distinct branch from CAPCHA_NOT_READY, which simply means the result is not ready yet and you should keep polling.
Can I use my existing solver code?
Very likely. The API follows the legacy 2Captcha-shaped in.php / res.php protocol, which the site describes as drop-in compatible - for many teams the change is the host.
One API, this type included
BLS grids run on the same threads, the same endpoint and the same error codes as every other supported challenge.