ILIndian Law 100
Concrete facts · verifiable law · auditable scoring

Legal AI India Benchmark

Legal AI India Benchmark measures how accurately an AI model applies Indian law to concrete facts. It contains 100 self-contained micro cases across 13 legal domains, 3 difficulty tiers, and deliberate traps for outdated or repealed law.

13 domains3 difficulty tiers10,000 points
Setup

Choose the model to test

Your key is sent only to OpenRouter. Save key stores it in this browser’s local storage; it is never included in benchmark results or sent to this website.

100 cases
Not saved
Provider groups and current per-million-token pricing will appear here.
Generation and scoring settings
Uses a separate model to grade every answer against the gold answer; choose “Use tested model” to make the tested model grade itself.
Controls randomness: lower values produce more consistent answers, while higher values allow more variation.
Limits token choices to the most likely probability mass; 1 keeps the full distribution available.
Sets the maximum length of each model answer; the judge uses a separate fixed limit.
Requests how much reasoning work a compatible model should perform before answering.
Reuses the same sampling starting point for better repeatability where the model supports seeds.

Recommended public-benchmark profile: temperature 0, Top P 1, 400 output tokens, model-default reasoning, and no seed. Any changed value is recorded in the exported JSON.

Loading the OpenRouter model catalogue…
No model selected · 100 cases
Add an API key and choose a model to begin
Running

The model is answering

Cases are isolated and sent one at a time. Each answer is scored immediately against its gold rubric.

0%0s elapsed0 / 100 cases
Preparing the first case…
Results

Auditable scorecard

Model answer, gold answer, and grade remain side by side. Open any grade to override the judge.

Overall0%
Points0 / 10,000
Correct0 / 100
Trap accuracy—
Estimated cost$0.0000
Case
Model's answer
Correct answer
Verdict

Finalising always downloads a JSON copy. A signed-in administrator can also save it to this server’s leaderboard. Checking administrator status…