Legal AI is a fast-growing vertical, dominated by closed proprietary platforms (Harvey, CoCounsel, Spellbook) for production deployment. Open-weight legal LLMs (SaulLM, Lawyer-LLaMA, LawGPT) cover research, on-prem privileged-data workflows, and components in larger legal AI pipelines. Quality on LegalBench and CaseHOLD benchmarks has improved materially in 2025-2026. This page consolidates the open-weight landscape.
Key Findings
- SaulLM from Equall is the leading open-weight legal LLM family, with 7B, 54B, and 141B variants demonstrating strong LegalBench performance.
- SaulLM 141B (released 2024) is the largest open-weight legal LLM, with approximately 81 percent LegalBench and approximately 74 percent CaseHOLD performance.
- Production legal AI overwhelmingly uses closed proprietary platforms (Harvey, CoCounsel, Spellbook, Lexis+, Westlaw Precision AI) for deployment; open-weight LLMs cover privileged-data on-prem workflows, internal automation, and components in larger pipelines.
- Legal hallucination remains the dominant deployment concern; production legal AI deployments require explicit citation grounding, retrieval over verified case databases, and human-in-the-loop review for case-bearing tasks.
- The 2026 production stack: general LLM (Claude 4.7 Opus, GPT-5.5) plus retrieval over verified legal corpora (Westlaw, Lexis, Bloomberg Law) plus citation validation; dedicated legal LLMs are typically components, not standalone solutions.
Open-Weight Legal LLM Comparison (May 2026)
| Model | Parameters | LegalBench | License |
|---|---|---|---|
| SaulLM 141B (Mistral-MoE base) | ~141B / ~39B active | ~81% | MIT |
| SaulLM 54B | ~54B | ~76% | MIT |
| SaulLM 7B | ~7B | ~62% | MIT |
| Lawyer-LLaMA 13B | ~13B | ~58% | Apache 2.0 |
| LawGPT | ~7B | ~54% | Apache 2.0 |
| Legal-BERT family | varies | n/a (classification model) | Apache 2.0 |
| Llama 3.1 70B (general, reference) | ~70B | ~71% | Llama 3.1 Community |
| Qwen3-235B-A22B Thinking (general) | ~235B MoE | ~78% | Apache 2.0 |
| Claude 4.7 Opus (closed reference) | n/a | ~86% | Closed |
| GPT-5.5 (closed reference) | n/a | ~87% | Closed |
Production Legal AI Platforms (Closed, For Reference)
| Platform | Status |
|---|---|
| Harvey | ~350+ law firms; flagship enterprise legal AI |
| CoCounsel (Thomson Reuters) | Westlaw integration |
| Spellbook (contract drafting) | Mid-market law firms |
| Lexis+ AI | LexisNexis integration |
| Westlaw Precision AI | Thomson Reuters research |
| Casetext / Allen & Overy ContractMatrix | Contract analysis |
| Eve (Khaira Law) | Plaintiff-side AI |
| Robin AI | Contract review |
Use Case Recommendations
| Use Case | Recommended Approach |
|---|---|
| General legal research | Harvey, CoCounsel, or Lexis+ AI (closed) for production |
| Privileged-data on-prem deployment | SaulLM 141B or Qwen3-235B-A22B Thinking with legal RAG |
| Contract review | Spellbook, Robin AI, or SaulLM with custom contract RAG |
| Case law research with citation grounding | General LLM with Westlaw or Lexis retrieval API |
| Compliance and regulatory monitoring | General LLM with regulatory text RAG |
| Patent search and analysis | SaulLM with USPTO patent corpus RAG |
| Cost-sensitive routine legal tasks | SaulLM 7B or 54B for self-hosted |
Hallucination and Production Concerns
Legal AI hallucination is well-documented and has resulted in court sanctions (multiple Mata v. Avianca-style cases through 2024-2026). Production legal AI deployments require: explicit citation grounding to verified case databases, retrieval validation that quoted text actually appears in the cited source, human-in-the-loop review for any case-bearing claim, and clear UX indicating "AI-generated; verify before relying" status. Open-weight legal LLMs are used as components in these production pipelines, not as standalone solutions.
Brand Visibility Implications
Legal AI is a high-citation enterprise procurement category. AI assistant queries about "legal LLM open source", "law firm AI tools", "SaulLM vs Harvey", and similar terms drive procurement-research traffic from law firms and corporate legal teams. Brands selling legal AI tools, contract review platforms, and law firm software face strong AI-mediated discovery surface for this category.
Methodology
Benchmark data compiled from primary model card disclosures, LegalBench and CaseHOLD evaluation publications, and the Hugging Face legal model leaderboard through 23 May 2026. Updated quarterly.
How Presenc AI Helps
Presenc AI monitors brand visibility on legal AI queries across ChatGPT, Claude, Gemini, and Perplexity. For legal AI tools, contract review platforms, and law firm software brands, the platform identifies the prompts driving procurement-research traffic and the gaps where new content unlocks share of voice.