Scale Labs PRBench Legal (PRBench Legal)
We show this table for reference; we do not rank on it.
A Scale Labs public leaderboard mirrored as display-only reference data. It does not affect BenchLM rankings.
Scale score on PRBench Legal — November 12, 2026 snapshot
We mirror the published scale score view for PRBench Legal. Muse Spark 1.3 leads the public snapshot at 61.6%, followed by Muse Spark Open (59.8%) and Muse Wonder 1.0 (58.4%). We do not use these results to rank models overall.
Muse Spark 1.3
Meta
Muse Spark Open
Meta
Muse Wonder 1.0
Meta
48 modelsKnowledgeCurrentDisplay onlyUpdated November 12, 2026 snapshot
Scale score table (48 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Muse Spark 1.3Meta | 61.6% | Not reported | Closed |
| 2 | Muse Spark OpenMeta | 59.8% | Not reported | Open |
| 3 | Muse Wonder 1.0Meta | 58.4% | Not reported | Unknown |
| 4 | Muse Spark 1.1Meta | 57.0% | Not reported | Closed |
| 5 | Gemini 4 CarbonGoogle | 53.7% | Not reported | Closed |
| 6 | claude-fable 5\n\nAnthropic | 52.6% | Not reported | Unknown |
| 7 | Muse SparkMeta | 52.3% | Not reported | Closed |
| 8 | claude-opus-4-6 (Non-Thinking)Anthropic | 52.3% | Not reported | Unknown |
| 9 | Fable 5.1 \nAnthropic | 51.6% | Not reported | Unknown |
| 10 | gpt-5.6-sol (max)\nOpenAI | 50.5% | Not reported | Unknown |
| 11 | gpt-5-proOpenAI | 49.9% | Not reported | Unknown |
| 12 | o3-proOpenAI | 49.7% | Not reported | Closed |
| 13 | Gemini 4 ArgonGoogle | 49.4% | Not reported | Closed |
| 14 | gpt-5.1-thinkingOpenAI | 49.3% | Not reported | Unknown |
| 15 | Claude Fable 5.2Anthropic | 49.0% | Not reported | Closed |
| 16 | gpt-5OpenAI | 49.0% | Not reported | Unknown |
| 17 | Gemini 3.8 FlashGoogle | 48.9% | Not reported | Closed |
| 18 | o3OpenAI | 48.6% | Not reported | Closed |
| 19 | GPT 6 Astra\nOpenAI | 48.4% | Not reported | Unknown |
| 20 | GPT-6.1 SolOpenAI | 47.9% | Not reported | Closed |
| 21 | Kimi K3p1 ProMoonshot AI | 47.9% | Not reported | Unknown |
| 22 | Grok 4.7 (xHigh)xAI | 47.6% | Not reported | Unknown |
| 23 | Claude Opus 5.5Anthropic | 47.6% | Not reported | Closed |
| 24 | Deepseek V4p1 Flash 1017Deepseek | 47.3% | Not reported | Unknown |
| 25 | GPT-5.2 ProOpenAI | 45.4% | Not reported | Closed |
| 26 | Minimax M3p1 FlashMiniMax | 45.1% | Not reported | Unknown |
| 27 | gpt-5.4 (High)OpenAI | 44.4% | Not reported | Unknown |
| 28 | Claude Opus 4.5 ThinkingAnthropic | 44.2% | Not reported | Closed |
| 29 | Gemini 3.1 ProGoogle | 44.0% | Not reported | Closed |
| 30 | Kimi K2.5Moonshot AI | 43.8% | Not reported | Open |
| 31 | GPT-6 LunaOpenAI | 42.9% | Not reported | Closed |
| 32 | Gemini 2.5 ProGoogle | 41.4% | Not reported | Closed |
| 33 | Gemini 2.5 FlashGoogle | 41.0% | Not reported | Closed |
| 34 | kimi-k2-thinkingMoonshot AI | 40.9% | Not reported | Unknown |
| 35 | Claude Sonnet 4.5Anthropic | 40.8% | Not reported | Closed |
| 36 | gemini-3-pro-previewGoogle | 40.6% | Not reported | Unknown |
| 37 | GPT-OSS 120BOpenAI | 40.2% | Not reported | Open |
| 38 | mistral-medium-latestMistral | 39.5% | Not reported | Unknown |
| 39 | GPT-6 SolOpenAI | 38.6% | Not reported | Closed |
| 40 | qwen.qwen3-235b-a22b-2507-v1:0Alibaba | 38.3% | Not reported | Unknown |
| 41 | o4-miniOpenAI | 38.1% | Not reported | Closed |
| 42 | deepseek-v3p1Deepseek | 37.6% | Not reported | Unknown |
| 43 | DeepSeek-R1-0528DeepSeek | 36.6% | Not reported | Open |
| 44 | GPT-4.1OpenAI | 36.5% | Not reported | Closed |
| 45 | kimi-k2-instructMoonshot AI | 36.4% | Not reported | Unknown |
| 46 | claude-opus-4-1-20250805Anthropic | 34.0% | Not reported | Unknown |
| 47 | GPT-4.1 miniOpenAI | 30.4% | Not reported | Closed |
| 48 | llama4-maverick-instruct-basicMeta | 24.8% | Not reported | Unknown |
How to read this leaderboard
Compare the published configurations as complete evaluation systems. The source can combine a base model, agent scaffold, tools, budget, and inference setting in each result.
Operator receipt: 48 sourced rows are currently displayable on this page; the leading published row is Muse Spark 1.3 at 61.6%.
Honest limit: This Scale table is display-only context, not benchmark provenance or a weighted model-only comparison.
How BenchLM shows PRBench Legal
BenchLM mirrors 48 published rows from Scale Labs’ public PRBench Legal leaderboard, captured on November 12, 2026 snapshot.
The table is display only. It is useful context for a published agent or model configuration, but it does not enter BenchLM’s overall or category rankings.
The published PRBench Legal snapshot places Muse Spark 1.3 first at 61.6%. The third row is 3.2 points behind. The broader top-10 range is 11.1 points, so the table still separates the published systems.
48 models have been evaluated on PRBench Legal. The benchmark falls in the Knowledge category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. PRBench Legal is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About PRBench Legal
Year
2026
Tasks
48 published rows
Format
Published Scale leaderboard score
Difficulty
External agent and model evaluation
BenchLM mirrors 48 published rows from the PRBench Legal public table captured on November 12, 2026 snapshot.
Freshness and provenance
Version
PRBench Legal 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does PRBench Legal measure?
A Scale Labs public leaderboard mirrored as display-only reference data. It does not affect BenchLM rankings.
Which model leads the published PRBench Legal snapshot?
Muse Spark 1.3 currently leads the published PRBench Legal snapshot with 61.6% scale score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on PRBench Legal?
The November 12, 2026 snapshot snapshot contains 48 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.