Guardrail evaluation on Representative guardrail task ~1250 tokens
0.01Evaluation Cost per 1KLuna-2
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Luna-2Model size=3B2026.02 | 0.01 | 150 | |
| GPT 4.1 mini2026.02 | 0.75 | 2,800 | |
| Azure Content Safety2026.02 | 1.65 | 312 | |
| GPT 4.12026.02 | 3.6 | 3,000 | |
| Gemini 3 ProThinking mode=low2026.02 | 5.85 | 6,900 | |
| Claude Sonnet 4.52026.02 | 8.2 | 6,700 |