BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

PRBench Legal

Professional reasoning in legal practice, graded by practitioners. Scale AI.

As of 24 Oct 2026, Muse Spark 1.3 leads PRBench Legal on BenchLeader with 61.6%, ahead of Muse Spark Open at 59.8%, across 42 model configurations with a published result.

Published by
Scale AI SEAL
Category
Knowledge
Index weight
0.5
Models
42
Data as of
24 Oct 2026
Newest result
20 Oct 2026

Scale AI SEAL Leaderboards.

What the test looks like

PRBench Legal poses professional legal tasks, from analysing a contract to advising on a scenario, graded by practising lawyers against rubrics.

How it is scored

Percent of rubric points achieved, published by Scale AI.

What to keep in mind

Jurisdiction-specific and rubric-graded; it complements the classification-style LegalBench.

Top 15
  1. 1Muse Spark 1.361.6%
  2. 2Muse Spark Open59.8%
  3. 3Muse Wonder58.4%
  4. 4Muse Spark 1.157.0%
  5. 5Claude Fable 552.6%
  6. 6Muse Spark52.3%
  7. 7Claude Opus 4.6 (none)52.3%
  8. 8Claude Fable 5.151.6%
  9. 9GPT-5.6 Sol (max)50.5%
  10. 10GPT-5 Pro49.9%
  11. 11o3-pro49.7%
  12. 12Gemini 4 Argon49.4%
  13. 13GPT-5.1 (thinking)49.3%
  14. 14GPT-549.0%
  15. 15Gemini 3.8 Flash48.9%
42 of 42
#
1Muse Spark 1.361.6%59.82026-09-14
2Muse Spark Open59.8%–2026-10-20
3Muse Wonder58.4%63.72026-10-14
4Muse Spark 1.157.0%62.52026-07-09
5Claude Fable 552.6%67.12026-07-28
6Muse Spark52.3%64.32026-04-08
7Claude Opus 4.6no reasoning52.3%56.52026-02-17
8Claude Fable 5.151.6%66.42026-09-03
9GPT-5.6 Solmax50.5%68.02026-07-28
10GPT-5 Pro49.9%59.82025-11-13
11o3-pro49.7%53.42025-11-13
12Gemini 4 Argon49.4%64.92026-10-16
13GPT-5.1thinking49.3%54.12025-12-11
14GPT-549.0%56.22025-11-13
15Gemini 3.8 Flash48.9%55.72026-09-09
16o348.6%59.52025-11-13
17GPT-6 Astra48.4%68.02026-09-09
18GPT-6.1 Sol47.9%63.02026-10-07
19Grok 4.7xhigh47.6%61.62026-09-25
20Claude Opus 5.547.6%62.02026-09-25
21GPT-5.2 Pro45.4%–2025-12-16
22GPT-5.4high44.4%57.92026-04-22
23Claude Opus 4.5thinking44.2%59.92025-11-26
24Gemini 3.1 Pro44.0%60.62026-03-23
25Kimi K2.543.8%54.52026-02-13
26GPT-6 Luna42.9%–2026-09-25
27Gemini 2.5 Pro41.4%53.92025-11-13
28Gemini 2.5 Flash41.0%52.72025-11-13
29Kimi K2 Thinkingthinking40.9%52.42025-11-13
30Claude Sonnet 4.540.8%53.72025-11-13
31Gemini 3 Pro40.6%60.42025-11-20
32gpt-oss-120b40.2%45.62025-11-13
33Mistral Medium 3.5medium39.5%41.12025-11-13
34GPT-6 Sol38.6%–2026-09-25
35Qwen3 235B A22B 250738.3%50.42025-11-13
36o4-mini38.1%56.92025-11-13
37DeepSeek R1 052836.6%49.52025-11-13
38GPT-4.136.5%48.32025-11-13
39Kimi K236.4%50.42025-11-13
40Claude Opus 4.134.0%51.52025-11-13
41GPT-4.1 mini30.4%45.02025-11-13
42Llama 4 Maverick Basic24.8%39.42025-11-13

More knowledge benchmarks · All LLM benchmarks

Cite as: BenchLeader, “PRBench Legal leaderboard”, data as of 24 Oct 2026, https://www.benchleader.com/benchmarks/scale_prbench_legal.

PRBench Legal: questions

What does PRBench Legal measure?
PRBench Legal poses professional legal tasks, from analysing a contract to advising on a scenario, graded by practising lawyers against rubrics. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads PRBench Legal?
Muse Spark 1.3 leads PRBench Legal with 61.6% as of 24 Oct 2026, ahead of Muse Spark Open at 59.8%.
How many models have PRBench Legal results?
42 model configurations have a PRBench Legal result on BenchLeader, all taken from Scale AI SEAL.
Who runs PRBench Legal and how often is it updated?
PRBench Legal is published by Scale AI SEAL. BenchLeader re-reads the published results once a day and records the date each result was published.
Does PRBench Legal count toward the BenchLeader Index?
Yes. PRBench Legal contributes to the knowledge category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.