Legal work
Legal reasoning, legal research and lawyer-style drafting. Some tests here check whether short answers are correct. Others check a research answer or a finished document against a checklist written for each task.
Sources last checked 2026-10-23.
We can't show scores for this kind of work yet. Vals AI publishes results from four tests. See who tests it
At a glance
Vendor claims have not been collected for any category yet.
ⓘ More about this category
- Reliability measured by a linked source; figures not shown here
- 4 sources, linked results only
- Sources last checked 2026-10-28
- Ring 4: needs rings 1 to 3 first
- LRB = Legal Research Bench, HLAB = HLAB-Vals-1.1 (Harvey's Legal Agent Benchmark), LGB = LegalBench, CL2 = CaseLaw v2
Tested by
Legal Research Bench
Vals AI, checked 2026-10-28Whether an AI agent with search of court decisions, web search and document tools answers US legal research questions with every point on a lawyer-written checklist.
Vals AI reports that models break down most consistently on questions that require weighing laws or court decisions that conflict.
Last updated 2026-10-26Covers 30 of the models we trackScores not shown: permission requested on 2026-10-24
See their resultsMore about this test: Legal Research Bench
Reliability not measured by this source.
Technical details for Legal Research Bench
All-Pass: Whether an AI agent with search of court decisions, web search and document tools answers US legal research questions with every point on a lawyer-written checklist.
License: Score reuse not stated. Checked 2026-10-28.
HLAB-Vals-1.1 (Harvey's Legal Agent Benchmark)
Vals AI, checked 2026-10-28Whether an AI agent working with files produces legal deliverables, such as documents, spreadsheets and slides, that meet every task-specific criterion.
Vals AI reports that the share of tasks fully resolved stays low even when models do well on individual criteria.
Last updated 2026-10-26Covers 8 of the models we trackHarvey, which sells AI software for legal and professional services, released this benchmark.Scores not shown: permission requested on 2026-10-24
See their resultsMore about this test: HLAB-Vals-1.1 (Harvey's Legal Agent Benchmark)
Reliability not measured by this source.
Technical details for HLAB-Vals-1.1 (Harvey's Legal Agent Benchmark)
Task Pass Rate: Whether an AI agent working with files produces legal deliverables, such as documents, spreadsheets and slides, that meet every task-specific criterion.
License: Score reuse not stated. Checked 2026-10-28.
LegalBench
Vals AI, checked 2026-10-28How often a model answers short legal reasoning tasks correctly, such as spotting legal issues, recalling rules, applying rules to facts and interpreting legal text.
Last updated 2026-10-13Covers 32 of the models we trackScores not shown: permission requested on 2026-10-24
See their resultsMore about this test: LegalBench
Reliability not measured by this source.
Technical details for LegalBench
Accuracy: How often a model answers short legal reasoning tasks correctly, such as spotting legal issues, recalling rules, applying rules to facts and interpreting legal text.
License: Score reuse not stated. Checked 2026-10-28.
CaseLaw v2
Vals AI, checked 2026-10-28How often a model correctly answers questions about recent Canadian court cases, such as finding the most relevant precedent or quoting the passage that answers a question.
Vals AI reports that many models struggle with the nuanced interpretation that accurate analysis of court cases requires.
Last updated 2026-05-04Covers 20 of the models we trackVals AI built this test with Jurisage, a legal technology startup whose AI helps law firms with case law and legal research.Scores not shown: permission requested on 2026-10-24
See their resultsMore about this test: CaseLaw v2
Reliability: Since February 2026 Vals runs this benchmark multiple times to reduce error.
Technical details for CaseLaw v2
Accuracy: How often a model correctly answers questions about recent Canadian court cases, such as finding the most relevant precedent or quoting the passage that answers a question.
License: Score reuse not stated. Checked 2026-10-28.
What this doesn't tell you
- None of these tests used your matters, your jurisdiction's local rules or your firm's templates.
- Check which country's law each test uses. CaseLaw v2 uses Canadian court cases.
- Scores from different sources are not comparable, even when they share a unit.
- None of these tests were run on your own documents, code or data.
- No source for this task allows us to show its figures yet. Follow the links for their results.
Data version 2026-10-28+91ac3fc9d071 · Terms of use