Grounding Hallucination Benchmark

RGB Closed-Book

RGB Closed-Book - row of the RGB hallucination-detection suite (closed-book setting) as printed on the Kolibri card in AA eval-framework setup.

Category

Grounding Hallucination

Score type

percent

Scale

100%

higher is better

Models tested

1

What is RGB Closed-Book?

RGB Closed-Book is a benchmark for whether answers stay grounded in provided sources. The official description says: "RGB Closed-Book - row of the RGB hallucination-detection suite (closed-book setting) as printed on the Kolibri card in AA eval-framework setup". Below we describe in simple words what benchmarks of this type generally examine.

What does it test?

RGB Closed-Book measures whether answers stay grounded in provided sources. In detail, it checks:

  • answering only from the given documents instead of inventing facts
  • admitting when the source does not contain the answer
  • correcting or flagging false statements in the source material
  • distinguishing between supported claims and plausible guesses

How to read the score

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

Official methodology note: RGB Closed-Book - row of the RGB hallucination-detection suite (closed-book setting) as printed on the Kolibri card in AA eval-framework setup.

Frequently asked questions

What is the RGB Closed-Book benchmark?

RGB Closed-Book is a benchmark for whether answers stay grounded in provided sources. The official description says: 'RGB Closed-Book - row of the RGB hallucination-detection suite (closed-book setting) as printed on the Kolibri card in AA eval-framework setup'. Below we describe in simple words what benchmarks of this type generally examine.

What does RGB Closed-Book test?

RGB Closed-Book mainly measures whether answers stay grounded in provided sources. Concretely, it checks: answering only from the given documents instead of inventing facts; admitting when the source does not contain the answer; correcting or flagging false statements in the source material.

How is the RGB Closed-Book score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get RGB Closed-Book data and scores?

The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/rgb-closed-book/scores) or pull it directly from the Open LLM Radar API with any HTTP client.

Model scores 1 model tested

Reset
# Model Score Date Source Verified
1 Kolibri-1
Aleph Alpha
51%
03.10.2026 Report Verified

What now?

You've seen how models score on RGB Closed-Book — here's what to do next.