Benchmark
HELM Lite: NaturalQuestions (open-book) - F1
Scores are percentages. Higher is better.
All knowledge benchmarksScores from Stanford HELM Lite data.
- Models measured
- 69
- Scores recorded
- 70
- Sources
- 1
- IQ weight
- 0.67
Score distribution
- 68.7%
- Mean
- 72.3%
- Median
- 12.3%
- Std dev
- 60.3%
- 10th percentile
- 77.6%
- 90th percentile
- 7.2%
- Min
This benchmark has a weight of 0.67 for chat models, measured over 69 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 | Phi-3 Small 8K Instruct· Stanford HELM Lite data | 67.5% |
| 52 | text-unicorn· Stanford HELM Lite data | 67.4% |
| 53 | Llama 2 70b· Stanford HELM Lite data | 67.4% |
| 54 | Llama-3.2-11B-Vision-Instruct-Turbo· Stanford HELM Lite data | 67.1% |
| 55 | Claude 2· Stanford HELM Lite data | 67% |
| 56 | gemma-7b· Stanford HELM Lite data | 66.5% |
| 57 | Open Mistral Nemo· Stanford HELM Lite data | 65% |
| 58 | Llama 2 13b· Stanford HELM Lite data | 64% |
| 59 | Claude Haiku 3.5· Stanford HELM Lite data | 63.9% |
| 60 | Jamba Instruct· Stanford HELM Lite data | 63.6% |
| 61 | Llama 2 7b· Stanford HELM Lite data | 61.2% |
| 62 | Claude 2.1· Stanford HELM Lite data | 61.1% |
| 63 | OLMo-7B· Stanford HELM Lite data | 60.3% |
| 64 | Mistral Small· Stanford HELM Lite data | 58.7% |
| 65 | Snowflake Arctic Instruct· Stanford HELM Lite data | 58.6% |
| 66 | DBRX-instruct· Stanford HELM Lite data | 55% |
| 67 | Mistral Large· Stanford HELM Lite data | 48.5% |
| 68 | Claude 3 Opus· Stanford HELM Lite data | 26.4% |
| 69 | Claude 3 Haiku· Stanford HELM Lite data | 25.2% |
| 70 | Claude 3 Sonnet· Stanford HELM Lite data | 7.2% |
Showing 51–70 of 70 results.
Data credits: Stanford HELM (Stanford CRFM), Apache-2.0 (licence)