Benchmark
hal-scicode
Coding benchmarks compare models on programming tasks. Scores are percentages.
All coding benchmarksScores from HAL (Princeton).
- Models measured
- 11
- Scores recorded
- 15
- Sources
- 1
- IQ weight
- 0.53
Score distribution
- 4.6%
- Mean
- 4.6%
- Median
- 3%
- Std dev
- 0%
- 10th percentile
- 9.2%
- 90th percentile
- 0%
- Min
This benchmark has a weight of 0.53 for chat models, measured over 11 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | |
|---|---|---|
| 1 | o4 Minilow· HAL (Princeton) | 9.2% |
| 2 | OpenAI o3medium· HAL (Princeton) | 9.2% |
| 3 | Claude Opus 4.1· HAL (Princeton) | 7.7% |
| 4 | Claude Opus 4.1high· HAL (Princeton) | 6.9% |
| 5 | GPT 4.1· HAL (Princeton) | 6.2% |
| 6 | GPT 5medium· HAL (Princeton) | 6.2% |
| 7 | o4 Minihigh· HAL (Princeton) | 6.2% |
| 8 | Claude Sonnet 3.7high· HAL (Princeton) | 4.6% |
| 9 | Claude Sonnet 3.7· HAL (Princeton) | 3.1% |
| 10 | Claude Sonnet 4.5· HAL (Princeton) | 3.1% |
| 11 | DeepSeek V3· HAL (Princeton) | 3.1% |
| 12 | Claude Sonnet 4.5high· HAL (Princeton) | 1.5% |
| 13 | Gemini 2.0 Flash· HAL (Princeton) | 1.5% |
| 14 | Claude Haiku 4.5· HAL (Princeton) | 0% |
| 15 | DeepSeek R1· HAL (Princeton) | 0% |
Tokens the model generated per task, as measured by the source. Comparable within this benchmark only.
Data credits: HAL (Princeton) — cite the HAL paper