Can AI solve JEE Advanced?

JEE Advanced Index

Latest AI performance on official English papers. Blind answers. Same IIT key for every model. 2026 peaks at 358 out of 360. 2025 is a clean 360 for Claude Fable 5.1.

Evaluations

198

7 model runs · 2 years

Goodmarks splits each official booklet into one PDF per question, sends that image to the model, and scores the JSON answer after the reply. The prompt never carries the key. Numerical ranges and multi-correct partial marks follow the paper’s own rules.

The 2026 field is four models. Claude Fable 5.1 leads on 358 after a published rescore on Paper 2 Mathematics question 4. On 2025, the same Fable run finishes 360 out of 360.

Showing JEE Advanced 2026

JEE Advanced 2026

2026 leaderboard

102 questions · 48 in Paper 1 · 54 in Paper 2 · held May 17, 2026

Peak 358 / 360Browse papers

2026 JEE Advanced Index

Marks out of 360 · Higher is better

4 of 4 models

1Claude Fable 5.1

Anthropic

358/ 360

2GPT 6 Astra

OpenAI

356/ 360

3GPT 5.5

OpenAI

351/ 360

4Claude Sonnet 5

Anthropic

341/ 360

Blind per-question scoring on official English papers. Official keys are applied after the model replies. 102 questions · held May 17, 2026.

2026 model comparison

Same paper, same key, same scoring rules

4 models

MetricClaude Fable 5.1AnthropicGPT 6 AstraOpenAIGPT 5.5OpenAIClaude Sonnet 5Anthropic
Index
JEE Advanced IndexHigher is better358356351341
Percent99.4%98.9%97.5%94.7%
Marks droppedLower is better24919
Papers
Paper 1 / 180180180180167
Paper 2 / 180178176171174
Subjects
Mathematics / 120120116116120
Physics / 120118120120118
Chemistry / 120120120115103
Accuracy
Correct10110110096
Partial1002
Wrong0122
Error0002
Report2026 write-up2026 write-upQuestion log2026 write-up

Paper × subject

Each cell is out of 60. Subjects are out of 120.

4 models

60 full55–5950–54under 50
ModelP1 MP1 PP1 CP2 MP2 PP2 CTotal
Claude Fable 5.1
60
60
60
60
58
60
358
GPT 6 Astra
60
60
60
56
60
60
356
GPT 5.5
60
60
60
56
60
55
351
Claude Sonnet 5
60
60
47
60
58
56
341

Where marks were lost

Paper 2 Mathematics question 4 is rescored on the written value. Our split image hid the A and B denominators. Models that wrote 1/3 get plus 3; a bare letter A stays minus 1.

10 misses

JEE Advanced 2025

2025 leaderboard

96 questions · 48 in Paper 1 · 48 in Paper 2 · held May 18, 2025

Peak 360 / 360Browse papers

2025 JEE Advanced Index

Marks out of 360 · Higher is better

3 of 3 models

1Claude Fable 5.1

Anthropic

360/ 360

2GPT 6 Astra

OpenAI

352/ 360

3GPT 5.6 Luna

OpenAI

319/ 360

Blind per-question scoring on official English papers. Official keys are applied after the model replies. 96 questions · held May 18, 2025.

2025 model comparison

Same paper, same key, same scoring rules

3 models

MetricClaude Fable 5.1AnthropicGPT 6 AstraOpenAIGPT 5.6 LunaOpenAI
Index
JEE Advanced IndexHigher is better360352319
Percent100%97.8%88.6%
Marks droppedLower is better0841
Papers
Paper 1 / 180180175167
Paper 2 / 180180177152
Subjects
Mathematics / 120120115107
Physics / 120120120110
Chemistry / 120120117102
Accuracy
Correct959385
Partial011
Wrong019
Error000
Report

Paper × subject

Each cell is out of 60. Subjects are out of 120.

3 models

60 full55–5950–54under 50
ModelP1 MP1 PP1 CP2 MP2 PP2 CTotal
Claude Fable 5.1
60
60
60
60
60
60
360
GPT 6 Astra
55
60
60
60
60
57
352
GPT 5.6 Luna
51
60
56
56
50
46
319

Where marks were lost

Paper 2 Physics question 4 is MARKS_TO_ALL on the official key, so every model receives full marks on that item.

12 misses

How we evaluate

Methodology

  1. Step 1

    Split the official papers

    Cut the English Paper 1 and Paper 2 booklets into single-question PDFs. Options, figures, and matching lists stay as printed.

  2. Step 2

    Keep every model blind

    Send only the question images and the type: single correct, multi correct, numerical, or matching. The official key never enters the prompt.

  3. Step 3

    Ask for a structured final answer

    Each model can think first. It has to finish with a JSON object. Scoring uses that final object, not the prose around it.

  4. Step 4

    Mark with the IIT key

    Apply official solutions after the reply. Use partial marks on multi correct items. Accept a numerical value only inside the published range.

Publications

Reports

The 2026 comparison write-up and the GPT 5.5 question log sit next to this board.

Practice

Practise the same exam shapes

Topic-wise JEE MCQs with instant checking and step-by-step solutions. Free samples on every topic. Pro unlocks the full bank.

Questions

Frequently asked questions

Can AI solve JEE?

Yes, on JEE Advanced. On official 2026 papers scored blind against the IIT key, Claude Fable 5.1 scored 358 out of 360 (99.4 percent). Claude Fable 5.1 358/360, GPT 6 Astra 356/360, GPT 5.5 351/360, Claude Sonnet 5 341/360. ChatGPT and Claude scores are on the Can AI solve JEE page.

What is the Goodmarks JEE Advanced benchmark?

A blind score of named frontier models on the official JEE Advanced English papers. Each question is sent as a printed image. Official keys are applied only after the model replies.

Who leads on JEE Advanced 2026?

Claude Fable 5.1 leads on 358 out of 360 (99.4 percent) after a published rescore on Paper 2 Mathematics question 4. GPT 6 Astra scored 356, GPT 5.5 scored 351, and Claude Sonnet 5 scored 341.

Who leads on JEE Advanced 2025?

Claude Fable 5.1 scored 360 out of 360. GPT 6 Astra scored 352. GPT 5.6 Luna scored 319.

Can ChatGPT solve JEE Advanced?

On official 2026 papers, GPT 6 Astra scored 356/360 and GPT 5.5 scored 351/360. On 2025, GPT 6 Astra scored 352/360 and GPT 5.6 Luna scored 319/360. The ChatGPT JEE score page lists every OpenAI miss.

Can Claude solve JEE Advanced?

Claude Fable 5.1 scored 358 out of 360 in 2026 and 360 out of 360 in 2025. Claude Sonnet 5 scored 341/360 in 2026. The Claude JEE score page lists every Anthropic miss.

Did the models see the official answers?

No. Each request included only the printed question and its type. Answers from the official solutions booklets were applied after the model replied.

Why was Paper 2 Mathematics question 4 rescored in 2026?

The split image hid the denominators under options A and B. Models that wrote the value 1/3 receive plus 3. A bare letter A stays minus 1. The write-up walks through the integral.

Can I see each question and the models’ reasoning?

Yes. Open a year, then a paper, then a subject, then a question. Each question page shows the printed image, the official key, and every model’s answer and reasoning. Misses on the leaderboard jump to that question.