Loading…
Every platform in this space claims students improve. We publish the numbers that let you verify. Updated quarterly.
All figures based on students with ≥6 mocks in the 45 days before the exam. Updated after each official cycle.
| Cycle | Cohort size | Metric | Result | Note |
|---|---|---|---|---|
| JEE Main Jan 2026 | 312 | Rank prediction accuracy | ±810 median · P50 hit within window for 78% of students | Based on students with ≥6 mocks in the 45 days before exam. |
| NEET UG 2025 (pilot) | 88 | Score improvement over 8-week cohort | +34 marks average vs. baseline mock | Pilot cohort. Controlled comparison group not available for this cycle. |
| JEE Main Apr 2026 | 541 | Rank prediction accuracy | ±790 median — improved from Jan cycle | Model retrained on Jan 2026 results. Cohort larger; confidence band narrowed. |
We do not cherry-pick or exclude outliers. All students who completed ≥6 mocks and gave us exam-day data are included. Next update: after NEET UG June 2026.
Publishing failures builds trust faster than hiding them. Most competitors won't do this. We will — every exam cycle, in this section.
In the Jan 2026 cycle, OCR confidence dropped below threshold on ~12% of grading submissions with very small or cramped handwriting. These were routed to human review (correct behaviour) but had a median 4-hour delay instead of our target <1 hour.
We retrained the OCR model on 2,400 additional handwriting samples from students in Tier 2/3 cities. Confidence threshold adjusted. As of March 2026, auto-grade rate is 94%.
Students in their second or third attempt showed higher prediction error than first-attempt students. Our model was undertrained on the drop-year cohort's score distribution.
Added drop-year flag to the prediction pipeline. Now trained separately on repeater vs. fresher cohorts. NEET prediction window for repeaters updated accordingly.
CUET domain subject score prediction was off by ±8 percentile points in our first attempt. CUET's marking scheme variability made it harder to calibrate against mock performance.
Now using subject-specific models for CUET instead of a single generalised model. Still improving — we'll publish updated accuracy after the June 2026 cycle.
Next update: After NEET UG June 2026 and JEE Advanced 2026 results are published. We commit to updating this page within 30 days of each official result.
We cannot claim causation. Students who use AspireACE more than 6 mocks may already be more motivated. Our improvement data does not have a randomised control group.
Our cohort is not a representative sample of all Indian exam aspirants. Early adopters skew urban and smartphone-fluent.
Prediction accuracy depends on mock quality. Students who game mocks or underperform intentionally will get worse predictions.
NEET data is thinner than JEE data. We need two more NEET cycles before we publish NEET accuracy with the same confidence as JEE.