Official marking scheme
Both UPSC Prelims papers deduct one-third of a question's marks for every wrong answer: 2/3 mark in GS Paper I and 5/6 mark in CSAT Paper II. Unanswered questions cost nothing. The rule comes from the “Penalty for wrong answers” note in Appendix-I of UPSC Examination Notice No. 05/2026-CSE.
Enter your correct and wrong answers to estimate your score after negative marking. Use 0 if you have none. No sign-in needed.
100 questions, 200 marks. Correct + wrong cannot exceed 100.
Enter both answer counts to calculate your score.
Uses the standard 2026 scheme of 100 GS or 80 CSAT questions. Dropped questions or revised answer keys may affect the official result. GS qualification also depends on the applicable cutoff.
Both papers are objective-type (MCQ). Each wrong answer deducts one-third of the marks assigned to that question.
| Paper | Questions | Total marks | Marks per question | Wrong answer | Unanswered |
|---|---|---|---|---|---|
| GS Paper I | 100 | 200 | 2 | −2/3 mark (about −0.67) | 0 |
| CSAT Paper II | 80 | 200 | 2.5 | −5/6 mark (about −0.83) | 0 |
For each wrong answer, one-third of the marks assigned to that question is deducted: 2/3 mark in GS Paper I and 5/6 mark in CSAT Paper II. Keep these fractions when calculating a score; round only the final total.
If you give more than one answer to a question, it is treated as a wrong answer and draws the same one-third penalty — even if one of the marked answers happens to be correct.
There is no penalty for a question you leave unanswered. A wrong GS answer loses 2/3 mark compared with a blank. The difference between a correct and a wrong answer is 8/3 marks.
GS Paper I and CSAT Paper II both carry the one-third penalty. CSAT requires at least 33%. This is 66 of the stated 200 marks. GS Paper I has a separate cutoff determined by UPSC.
Worked example
Say you attempt 80 GS questions and get 55 right and 25 wrong. You earn 55 × 2 = 110 marks and lose 25 × 2/3 marks. Your net score is 93.33 when rounded to two decimal places. Compare it with the relevant year and category in the official cutoffs.
The same 80 attempts with 45 right and 35 wrong gives 90 − 35 × 2/3 = 66.67 marks after rounding. Use the exact penalty for each answer, then round the final score.
When guessing pays
A blind guess across four options expects 0.25 × 2 − 0.75 × 2/3 = 0 expected marks. If you correctly eliminate one option and choose equally among the remaining three, the expected gain is about 0.22 marks. Correctly eliminate two and it rises to about 0.67 marks.
These averages assume your eliminated options are wrong. If you eliminate the correct answer, the calculation no longer applies. Consider your confidence and remaining time; a positive expected value does not guarantee marks on an individual question.
GS Paper I decides the Prelims cutoff and who advances to Mains. CSAT Paper II is qualifying only at 33%, or 66 of 200 marks. Prelims marks do not count toward the final merit list; final merit is based on Mains and Interview.
That makes net score after penalties the only number that matters. Practise with timed mock tests that apply the same one-third deduction, so you can compare practice attempts using the same scoring rule.
The rules below come from pages 25 and 26 of the 2026 notice. The worked scores and guessing probabilities are our calculations using its one-third penalty. Check the relevant year's notice and final answer key when estimating an actual result.
Appendix-I: the two-paper scheme, the CSAT 33% qualifying rule, and the 'Penalty for wrong answers' note that defines the one-third deduction.
The full two-paper structure: questions, marks, duration, and qualifying rules.
Open resource →Timed mocks scored with the same one-third penalty, so your practice score is honest.
Open resource →Official year-wise cutoffs — see what a safe net score looks like after negative marking.
Open resource →Practise real UPSC questions and learn when to attempt, eliminate, or skip.
Open resource →Timed mock tests scored with one-third negative marking, PYQ drills, and accuracy analytics.