AI benchmarks and leaderboards are different things
Learn the difference between a benchmark, an evaluation run and a leaderboard, and use each to answer the right question.
LLM Stats
XSTest
XSTest is a test suite designed to identify exaggerated safety behaviours in large language models. It comprises 450 prompts: 250 safe prompts across ten prompt types that well-calibrated models should not refuse to comply with, and 200 unsafe prompts as contrasts that models should refuse. The benchmark systematically evaluates whether models refuse to respond to clearly safe prompts due to overly cautious safety mechanisms.
Evaluation context and evidence
- Source
- LLM Stats
- Publisher
- Not recorded
- Benchmark release
- Not recorded
- Collected
- 2026-08-17
- Modality
- text
These are source-reported observations, not independent reruns by this site. Check the task version, evaluation protocol, unit and reporting date before comparing models. Results from different setups are not a universal ranking.
Original sources
Reported results 4
Shown in source observation order, not ranked. Dates retain the precision provided by the source.
| Model | Value | Unit | Protocol | Reported date and basis | Source |
|---|---|---|---|---|---|
| Gemini 1.5 Pro | 0.988 | Not recorded | Not recorded | 2024-05-01 · model_announcement | LLM Stats |
| Gemini 1.5 Flash | 0.97 | Not recorded | Not recorded | 2024-05-01 · model_announcement | LLM Stats |
| Shieldstral 1.0 (3B) | 0.946 | Not recorded | Not recorded | 2026-08-04 · model_announcement | LLM Stats |
| Gemini 1.5 Flash 8B | 0.926 | Not recorded | Not recorded | 2024-03-15 · model_announcement | LLM Stats |