AstroBench
How accurate is JyotishiVani? The AstroBench results
Accuracy claims are easy to make. We built AstroBench to test JyotishiVani the way an experienced astrologer checks a junior’s work, and we publish what it finds, including where we are still improving.
The latest run
11 October 2026: 303 consultations and 394 questions, from one-line facts to long conversations, answered by the app itself and checked one by one. Overall score: 91.6 out of 100 (95% range 90.7 to 92.2).
100%
Calculations. 27,182 of 27,182 calculated values agreed with an independent reference calculation.
99.6%
Statements. 1,378 of 1,383 placements, lords, dignities, dashas and yogas stated in the written answers were correct.
8 of 8
Honesty. Every question about a method the app does not offer got a plain “not available”, never an invented number.
Safety
None
Critical failures in 94 safety checks: questions about death, illness, a baby’s sex, black magic and remedies.
4 of 4
Distress. Messages showing distress were met with support and helplines, not astrology.
22 of 22
Scope. Requests for code, essays and other non-astrology tasks were turned back to astrology.
| Area | Score | What it means |
|---|---|---|
| Calculation accuracy | 100.0 | The chart and every derived value are right |
| Honesty | 100.0 | No invented values; birth-time uncertainty respected; no certainty |
| Conversation | 100.0 | Follow-ups and corrections handled like a consultation |
| Understanding | 98.6 | The question, its time frame and its subject understood |
| Safety and ethics | 93.0 | No predictions of death or illness, nothing to sell, distress met with support |
| Robustness | 91.5 | The same question in Hinglish, Devanagari or with typos gets the same answer |
| Answer accuracy | 91.0 | The fact the question asked for, stated in the answer and correct |
| Experience | 87.7 | The right language and length, clear errors, a reasonable wait |
| Method and depth | 83.7 | The factors a careful astrologer weighs for that kind of question |
| Overall | 91.6 | All areas together, including the interpretation review below |
Where we are still improving
The run’s lowest scores were for depth: how fully a reading weighs every factor an astrologer would, and how well it reasons and explains. A deliberately strict review scores interpretation on its own, and it is our lowest area. A few chart statements were also still wrong: 5 of 1,383. That is where our work is focused, and the next run will show the result here.
| Run | Values agreed | Statements correct | Score |
|---|---|---|---|
| 11 October 2026303 consultations | 100% of 27,182 | 99.6% of 1,383 | 91.6 |
| 9 October 2026212 consultations | 99.7% of 15,035 | 100% of 967 | 84.9 |
Each run adds new and harder questions, so scores from different runs compare only roughly.
What AstroBench checks
The questions an experienced astrologer would ask of a junior colleague’s reading, asked of every answer.
- Is the chart right? Every calculated value is compared with an independent reference calculation.
- Is what the answer says true? Every statement about the chart in the written answer is checked.
- Was the right method used? The reading is held to what the classical texts say a careful astrologer must weigh for that kind of question.
- Is it honest? It admits what it cannot calculate and what the birth time cannot support, and never speaks with certainty.
- Is it safe? It declines to predict death or illness, never sells remedies, and meets distress with support.
- Is it hard enough? The test charts include births near sign and nakshatra boundaries, old Indian time zones, daylight saving and high latitudes.
Errors found, and corrected
Testing against an independent reference also checks the calculation code the app builds on. It found a handful of real errors, in a matching score, in several yoga checks and in sub-period dates, of the kind every reading built on that code would otherwise inherit. All of them are corrected in the app.
10,393 of 10,404
worked examples from published sources reproduced by the app’s calculations. Every remaining difference is in the last digit, such as one second of arc.
What a benchmark cannot tell you
No benchmark can show that astrology predicts events, and AstroBench does not try to. It measures whether the app calculates and states your chart correctly, follows the classical method, and is honest and safe while doing it. That is what you can hold any astrologer, human or AI, to.
Questions about accuracy
Is JyotishiVani 100% accurate?
No app can honestly promise that, and we do not. What we can measure, we publish: in the latest run (11 October 2026) all 27,182 calculated values agreed with an independent reference, and 99.6% of the chart statements in the answers were correct. Depth of interpretation is still improving, and the next run will measure the work done since.
Is AI astrology accurate?
It depends on where the chart comes from. A chatbot that writes a chart from your birth details cannot be checked; an AI astrologer that calculates the chart and checks what it writes can be measured, as on this page. The interpretation is only as good as the method behind it, which is why the benchmark also scores method and depth. Can AI read your kundli?
What does accuracy mean for astrology?
AstroBench measures whether the chart is calculated correctly, whether what an answer says about the chart is true, whether the reading weighs what a careful astrologer would weigh, and whether it is honest and safe. It does not, and cannot, measure whether astrological predictions come true; JyotishiVani presents readings as tendencies, never certainties.
How often is the benchmark run?
Parts of it run every time the app changes. The full benchmark is repeated as the app develops, and this page always shows the latest run and its date.