Sovereign Bench
Benchmark Leaderboard Methodology Account
◆

Sign in to run the benchmark

You need an account to run Sovereign Bench. Your results are saved automatically and you can submit them to the public leaderboard.

Sign In / Sign Up View Leaderboard

Free accounts can run benchmarks, use the API, and download results. Upgrade to Sovereign for higher API limits, priority scoring, and permanent storage.

Run the Benchmark

Configure your benchmark run. You will be guided through all prompts one at a time. Copy each prompt into your model, paste the response back.

Enable only when you can supply the tested response’s visible reasoning text. Model capability alone does not establish trace availability. Enables IOCI; a summary is not the complete internal reasoning.

Higher difficulties add more prompts per axis. All difficulties include the standard set.

Leave blank to submit anonymously.

Prompt 1 of 38

OPERATOR RESPECT

Axis 1a: Sycophancy Detection

Does the model validate incorrect claims to preserve the user's face?

⇄
Turn 1 This is a multi-turn prompt. Keep the same conversation going in your model — do not start a new chat.
PROMPT
Loading prompt...

Use only reasoning or summary text actually exposed with this response; it may be incomplete and does not reveal all internal reasoning. This is scored on the IOCI axis and used to evaluate coherence between internal reasoning and external output across all axes.

Benchmark Complete

All responses collected. Review the summary below, then submit for scoring.

Scoring v1.04 uses a provisional equal-weight judge panel served through Together AI. Judges receive the required context for each rubric-bearing response. Every selected judge must return a valid score.

Home Methodology Leaderboard Privacy Terms API Docs Changelog Report a Bug
Kuykendall Industries LLC

Report a Bug

0/5000

Thank you!

Your report has been submitted. We'll look into it.