PRBench Legal leaderboard
- Muse Spark 1.3 leads
- 61.6%
- models measured
- 35
- lead over #2
- +1.8 pts
- median
- 47.6%
Leader on PRBench Legal · read October 22
On October 22, 2026, 35 of our catalog's models had a PRBench Legal result, and Meta Muse Spark 1.3 led with 61.6%, ahead of Meta Muse Spark Open at 59.8%.
PRBench Legal tests complex legal reasoning tasks drawn from real legal practice and case analysis. The number is the score Scale AI publishes for each model.
Role in the index
PRBench Legal is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Muse Spark 1.3Meta | 61.6% | 57.4 | $4.25 | |
| 2 | Muse Spark OpenMeta | 59.8% | 47.4 | $4.25 | |
| 3 | Muse Spark 1.1Meta | 57.0% | 70.3 | $4.25 | |
| 4 | Claude Fable 5Anthropic | 52.6% | 79.3 | $50.00 | |
| 5 | Claude Opus 4.6Anthropic | 52.3% | 57.2 | $25.00 | |
| 6 | Claude Fable 5.1Anthropic | 51.6% | 84.2 | $50.00 | |
| 7 | GPT-5.6 SolOpenAI | 50.5% | 73.5 | $20.00 | |
| 8 | GPT-5 ProOpenAI | 49.9% | Not benchmarked | $120.00 | |
| 9 | o3 ProOpenAI | 49.7% | Not benchmarked | $80.00 | |
| 10 | Gemini 4 ArgonGoogle | 49.4% | 81.0 | $10.00 | |
| 11 | GPT-5.1OpenAI | 49.3% | 40.4 | $10.00 | |
| 12 | GPT-5OpenAI | 49.0% | 38.8 | $10.00 | |
| 13 | Gemini 3.8 FlashGoogle | 48.9% | 61.6 | $3.75 | |
| 14 | o3OpenAI | 48.6% | Not benchmarked | $8.00 | |
| 15 | GPT-6 AstraOpenAI | 48.4% | 85.1 | $50.00 | |
| 16 | GPT-6.1 SolOpenAI | 47.9% | 82.4 | $10.00 | |
| 17 | Grok 4.7xAI | 47.6% | 66.7 | $6.00 | |
| 18 | Claude Opus 5.5Anthropic | 47.6% | 88.8 | $20.00 | |
| 19 | GPT-5.2 ProOpenAI | 45.4% | Not benchmarked | $168.00 | |
| 20 | GPT-5.4OpenAI | 44.4% | 64.4 | $15.00 | |
| 21 | Claude Opus 4.5Anthropic | 44.2% | 44.0 | $25.00 | |
| 22 | Kimi K2.5Moonshot AI | 43.8% | 38.6 | $2.85 | |
| 23 | GPT-6 LunaOpenAI | 42.9% | 57.3 | $0.50 | |
| 24 | Gemini 2.5 ProGoogle | 41.4% | 19.1 | $10.00 | |
| 25 | Gemini 2.5 FlashGoogle | 41.0% | Not benchmarked | $2.50 | |
| 26 | Kimi K2 ThinkingMoonshot AI | 40.9% | Not benchmarked | $2.50 | |
| 27 | Claude Sonnet 4.5Anthropic | 40.8% | 39.5 | $15.00 | |
| 28 | gpt-oss-120bOpenAI | 40.2% | Not benchmarked | $0.50 | |
| 29 | GPT-6 SolOpenAI | 38.6% | 76.0 | $10.00 | |
| 30 | o4 MiniOpenAI | 38.1% | Not benchmarked | $4.40 | |
| 31 | R1 0528DeepSeek | 36.6% | Not benchmarked | $2.18 | |
| 32 | GPT-4.1OpenAI | 36.5% | Not benchmarked | $8.00 | |
| 33 | Kimi K2 0711Moonshot AI | 36.4% | Not benchmarked | $2.30 | |
| 34 | Claude Opus 4.1Anthropic | 34.0% | Not benchmarked | $75.00 | |
| 35 | GPT-4.1 MiniOpenAI | 30.4% | Not benchmarked | $1.60 |
Only models in our catalog are listed, and each appears once with its best published run. 66 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out.