Benchmark
hal-swe-bench-verified-mini
Coding benchmarks compare models on programming tasks. Scores are percentages.
All coding benchmarksScores from HAL (Princeton).
- Models measured
- 12
- Scores recorded
- 18
- Sources
- 1
- IQ weight
- 1.61
Score distribution
- 44.5%
- Mean
- 50%
- Median
- 16.5%
- Std dev
- 24%
- 10th percentile
- 61%
- 90th percentile
- 6%
- Min
This benchmark has a weight of 1.61 for chat models, measured over 12 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 | Claude Sonnet 4.5high· HAL (Princeton) | 72% |
| 2 | Claude Sonnet 4.5· HAL (Princeton) | 68% |
| 3 | Claude Opus 4.1· HAL (Princeton) | 61% |
| 4 | Claude Opus 4.1high· HAL (Princeton) | 54% |
| 5 | Claude Sonnet 3.7high· HAL (Princeton) | 54% |
| 6 | o4 Minilow· HAL (Princeton) | 54% |
| 7 | Claude Opus 4· HAL (Princeton) | 50% |
| 8 | Claude Sonnet 3.7· HAL (Princeton) | 50% |
| 9 | o4 Minihigh· HAL (Princeton) | 50% |
| 10 | GPT 5medium· HAL (Princeton) | 46% |
| 11 | OpenAI o3medium· HAL (Princeton) | 46% |
| 12 | Claude Haiku 4.5high· HAL (Princeton) | 44% |
| 13 | GPT 4.1· HAL (Princeton) | 44% |
| 14 | Claude Opus 4high· HAL (Princeton) | 30% |
| 15 | Claude Haiku 4.5· HAL (Princeton) | 24% |
| 16 | DeepSeek V3· HAL (Princeton) | 24% |
| 17 | Gemini 2.0 Flash· HAL (Princeton) | 24% |
| 18 | DeepSeek R1· HAL (Princeton) | 6% |
Tokens the model generated per task, as measured by the source. Comparable within this benchmark only.
Data credits: HAL (Princeton) — cite the HAL paper