73.4%
#1
CaseLaw v2 accuracy of GPT 5.1, the highest of the top 15; GPT 4.1 scored 69.9%
What the chart shows
CaseLaw v2 tests models on litigation and legal research tasks built from recent US and Canadian court decisions, mostly dated after models’ training cut-offs, so they must reason over the documents supplied. GPT 5.1 led at 73.4%, followed by GPT 4.1 (69.9%) and GPT 5 Mini (68.5%).
The rest of the top 15 fall between 62.1% and 66.5%, with Claude Opus 4.6 (Thinking) lowest. The AI Index notes a recurring weakness: models lean on general legal knowledge instead of the supplied documents, even when told not to.
Key numbers
| Measure | Value |
|---|---|
| GPT 5.1 | 73.42% |
| GPT 4.1 | 69.88% |
| GPT 5 Mini | 68.49% |
| GPT 5 | 66.45% |
| Claude Opus 4.6 (Thinking), 15th | 62.06% |
| Gap, GPT 5.1 to GPT 4.1 | 3.54 percentage points |
How to read it
- The benchmark has 300 validation and 104 test questions across seven kinds of legal reasoning, per the AI Index.
- Scores are from Vals.ai’s CaseLaw v2 leaderboard; the AI Index shows the top 15 models.
Source: Vals.ai, 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.5.12: CaseLaw v2: accuracy. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion