Benchmark
Phare Hallucination Resistance
Factuality benchmarks measure how often models get facts right. Scores are percentages.
- Models measured
- 70
- Scores recorded
- 70
- Sources
- 1
- IQ weight
- 1.41
Score distribution
- 74.1%
- Mean
- 75.2%
- Median
- 8.3%
- Std dev
- 63.2%
- 10th percentile
- 85.5%
- 90th percentile
- 53.9%
- Min
This benchmark has a weight of 1.41 for chat models, measured over 70 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 51 | Gemini 2.5 Flash-Lite· Phare | 68.1% |
| 52 | Mistral Large 3· Phare | 68% |
| 53 | GPT-4.1 mini· Phare | 67.8% |
| 54 | GPT-4.1 nano· Phare | 67.7% |
| 55 | Mistral Medium 3.5· Phare | 67.7% |
| 56 | Llama 3.1 405B Instruct· Phare | 67.4% |
| 57 | Grok 2· Phare | 66.8% |
| 58 | GPT OSS 120B· Phare | 66.1% |
| 59 | GPT-4o mini· Phare | 66% |
| 60 | Llama 3.1 8B Instruct· Phare | 63.8% |
| 61 | Grok 4 Fastnon-reasoning· Phare | 63.4% |
| 62 | Mistral Small2506· Phare | 63.4% |
| 63 | Gemini 2.0 Flash Lite· Phare | 63.2% |
| 64 | Magistral Small 2509latest· Phare | 62.4% |
| 65 | DeepSeek V3.1· Phare | 61.6% |
| 66 | Qwen3 8B· Phare | 60.4% |
| 67 | Gemma 3 27B IT· Phare | 60.3% |
| 68 | Qwen3 30B A3B Instruct· Phare | 59.9% |
| 69 | Llama 4 Scout· Phare | 58.2% |
| 70 | Gemma 3 12B IT· Phare | 53.9% |
Showing 51–70 of 70 results.
Data credits: Phare (Giskard), CC BY 4.0 (licence)