Which GPT Model Is Best for Legal Work? What Lawyers Found Testing GPT-6.1 Sol and GPT-6.1 Terra
TL;DR
- Drafting and quick turnaround: GPT-6.1 Terra for emails, short memos, first-pass clauses, and well-scoped document work where speed matters. It posted the best drafting result of any OpenAI model the team has evaluated in Lito, and it completes tasks markedly faster than GPT-6.1 Sol
- Summaries and harder reasoning: GPT-6.1 Sol for summarization, extraction, and the harder analysis and reasoning where Terra gives ground. It is among the strongest OpenAI models the team has evaluated in Lito
- Questions about what the law is: give either model the source material. Both were weakest when a question turned on general legal knowledge rather than a document in front of them
The gap between the two is narrow. For both, a precise instruction on scope and length is the most effective way to get a better answer. These findings come from the Litera Legal Knowledge Engineering (LKE) team, made up of former practicing lawyers who tested each model on real legal work in Lito, Litera’s award-winning Legal AI agent.
How Did Litera Test GPT-6.1 Sol and GPT-6.1 Terra for Legal Work?
The LKE team tested each model out of the box, on real legal work in Lito. The team is made up of former practicing lawyers across multiple jurisdictions and practice areas.
The goal was not to rank models on generic benchmarks. It was to describe what each model is like to work with as lawyers will encounter it today, and when to reach for which. The team assessed each model’s default behavior at a point in time. Model behavior evolves, and individual results will vary with prompts and documents.
OpenAI positions GPT-6.1 Sol, an upgrade to GPT-6 Sol, as near-Astra performance for complex professional work at a lower cost. It positions GPT-6.1 Terra, released a month later, as a fast, efficient model for clearer, more natural writing, visual reasoning, and computer use, and a significant step up from GPT-5.6 Terra. Both models are now available in Lito.
Which GPT Model Should Lawyers Use?
Both models are among the strongest OpenAI models the LKE team has evaluated in Lito, and Terra sits close behind Sol overall. Where they differ is in the shape of their strengths and in speed.
| Model | Reach for it when | Strongest at | Watch for | Speed and cost |
|---|---|---|---|---|
| GPT-6.1 Terra | The task is drafting or well-scoped document work and quick turnaround matters | Drafting: the best result of any OpenAI model evaluated in Lito, by an edge rather than a clear lead | A gap to Sol on applied legal reasoning, expert judgment, and analysis. Weakest on general legal knowledge | Completes tasks markedly faster than Sol, at the lower cost of the two |
| GPT-6.1 Sol | The work is summarization or extraction, or the analysis and reasoning is harder | Summarization: the best of any OpenAI model evaluated. Direct recall: tied for the strongest. Biggest gains on expert judgment, applied legal reasoning, and analysis | Weakest on questions about what the law is. Analysis and drafting trail its summarization and extraction | Slower than Terra. OpenAI positions Sol as faster and lower cost than its top tier |
Is GPT-6.1 Sol Good for Legal Summaries and Complex Analysis?
Yes. GPT-6.1 Sol posted among the strongest results of any OpenAI model the LKE team has evaluated in Lito. It is a clear step up from GPT-6 Sol, the model it replaces, and it improved on GPT-5.6 Sol, the previous generation of the Sol line.
Where it leads: The gains were concentrated in the hardest work rather than spread evenly. The largest came on expert judgment, applied legal reasoning, and analysis, the tasks where earlier OpenAI models had most often fallen short. Sol also posted the best summarization result of any OpenAI model the team has evaluated, and it tied for the strongest on direct recall. Summarizing a lease’s insurance requirements, Sol captured virtually all of the substantive requirements in a concise format that evaluators found easy to review, and it added accurate detail drawn from the source.
Style: By default, Sol’s answers ran to about the same length as GPT-5.6 Sol’s, with noticeably fewer bullet points, so the content arrives as readable prose with structure where it earns its place. Sol also cited the source documents more densely than its predecessor and rarely left a document-based answer without a citation. When a question spanned several documents, it drew on each of them. For work where a lawyer needs to check every proposition against the source, that habit saves time.
Where it gives ground
- Questions about what the law is: Sol was weakest on questions that turn on what the law is rather than what a document says
- Analysis and drafting: despite its gains, results here were less strong than on summarization and extraction, and on drafting GPT-6.1 Terra held a slight edge
For a client-facing draft on a difficult point, give the first response a closer read before it goes further.
Is GPT-6.1 Terra Good for Legal Drafting?
Yes. Drafting is where Terra stood out. It posted the best drafting result of any OpenAI model the LKE team has evaluated in Lito, ahead of GPT-6.1 Sol by a margin best read as an edge rather than a clear lead. Drafting is also where Terra moved furthest beyond GPT-5.6 Terra. That fits OpenAI’s emphasis on clear, natural writing, and it makes Terra a good first stop for an email to a client, a short memo, or a first pass at a clause.
Speed and cost: Overall, Terra came in behind Sol in the team’s testing, but the gap was narrow. It completed tasks markedly faster than Sol, and it is the lower-cost tier of the two. For well-scoped work where turnaround matters, that combination is the reason to choose it.
Style: By default, Terra’s answers were shorter than Sol’s and lightly formatted, with fewer headings and bullet points and only an occasional table, so they read as prose rather than as a briefing. Terra also cited its sources densely, more so than Sol, and almost never left a document-based answer without a citation in Lito. If a lawyer wants a table or a structured summary, the lawyer can ask for one.
Where it gives ground
- Hardest reasoning: the gap to Sol opened on applied legal reasoning, expert judgment, and analysis, in line with OpenAI positioning Sol above Terra for complex work
- General legal knowledge: Terra was weakest of all on questions that turn on general legal knowledge rather than a source in front of it
How Should Lawyers Prompt AI Models for Legal Work?
GPT-6.1 Sol and GPT-6.1 Terra are among the strongest OpenAI models the LKE team has evaluated in Lito. They still carry the same limitations as any large language model (LLM), and the same good habits apply. Three prompting habits improve results from either model:
- Be explicit about scope: longer answers from either model were no more likely to be right than shorter ones. When a task calls for a narrow, bounded deliverable, say so directly, and name what to exclude, not only what to include
- State your jurisdiction: when a task does not specify one, models tend to default to US law without flagging the assumption
- Ground it in current sources: both models were weakest on questions that turn on general legal knowledge rather than a document in front of them. Where a task turns on the current state of the law, provide source material rather than relying on the model’s recall
How Should Lawyers Review AI Output on Legal Work?
Three review habits still apply, however strong the model:
- Check for the missing point: both models satisfied nearly every individual point a legal answer needs, yet fell short of a fully complete answer far more often. The gap was usually missed content rather than wrong content, which makes it easy to overlook. Summarizing a lease’s use restrictions, Sol captured the general approval standard but missed specific restrictions set out elsewhere in the lease. Read the first response as a strong draft, check it against every part of the original ask, and use a follow-up prompt to fill any gap
- Take more care as the task gets harder: performance is consistently strong on locating, extracting, and summarizing information, and drops for every model as tasks move into applied legal reasoning and expert judgment. The more a task depends on judgment, the more closely its conclusions deserve review
- Verify assertions on nuanced points of law: confirm that legal conclusions are grounded in the cited source rather than general knowledge
Why Does Model Choice Alone Not Solve the Problem?
Because a strong model is still a probabilistic one. The LKE findings point to the same conclusion for any model a firm picks: even the strongest need grounding, structure, and review on legal work, and an answer that varies from one run to the next presents too much risk where the work cannot be wrong. AI-first entrants can look compelling in a model demo and underdeliver in deployment. A strong foundation, not a bold promise, is the better bet.
That foundation is what Litera has built for 30 years. Litera is trusted Legal AI that best unifies the practice and business of law, and Lito is how lawyers put that platform to work. Lito combines leading foundational models, including GPT-6.1 Sol and GPT-6.1 Terra, with rules-based engines, skills curated by Litera in-house lawyers, and the firm’s own data. It runs inside Word, Outlook, on the web, and on iOS, so lawyers do not need a new platform or a new habit.
What Is the Return on Using the Right Model in Lito?
Choosing the right model for each task, and grounding it in deterministic engines where accuracy matters most, is how Litera delivers the full Return on AI (RoAI), not efficiency alone. Firms that invest in efficiency without an equal focus on growth risk optimizing toward lower revenue. Only Litera is built to deliver all three components:
- Efficiency Growth: the right model for each task means faster turnaround at lower cost, without paying for capability the task does not need
- Relationship Growth: answers grounded in exact results, and reviewed with the right habits, give clients work they can trust
- Business Growth: time returned to lawyers becomes capacity for more matters, and firm intelligence helps turn existing relationships into new work
See How Lito Puts the Right Model to Work
The best way to judge any Legal AI is on the work your lawyers do every day. GPT-6.1 Sol and GPT-6.1 Terra are available in Lito now. Explore Lito, or let’s talk about how your team can match the right model to each task.
Frequently Asked Questions About GPT Models and Lito
Which AI model is best for legal work?
It depends on the task. GPT-6.1 Terra is the stronger pick for drafting and for well-scoped work where turnaround matters. GPT-6.1 Sol is the stronger pick for summaries, extraction, and harder analysis and reasoning. For questions about what the law is, give either model the source material.
Which GPT model is best for legal drafting?
GPT-6.1 Terra. In LKE testing, it posted the best drafting result of any OpenAI model evaluated in Lito, ahead of GPT-6.1 Sol by an edge rather than a clear lead.
Which GPT model is best for summarizing legal documents?
GPT-6.1 Sol. It posted the best summarization result of any OpenAI model the LKE team has evaluated, and it tied for the strongest on direct recall.
Which GPT model is best for complex legal analysis?
GPT-6.1 Sol. The gap to Terra opened on applied legal reasoning, expert judgment, and analysis, where Sol made its largest gains over earlier OpenAI models.
Which GPT model is faster?
GPT-6.1 Terra. It completed tasks markedly faster than Sol in LKE testing, and it is the lower-cost tier of the two.
Which GPT model is best for questions about what the law is?
Neither is strongest on its own. Both models were weakest on questions that turn on general legal knowledge rather than a document in front of them, so provide the source material.
Do AI models assume US law?
Often, yes. When a task does not specify a jurisdiction, models tend to default to US law without flagging the assumption. State your jurisdiction in the prompt.
Can lawyers rely on AI output as final work product?
Treat the first response as a strong draft. Both models satisfied nearly every individual point a legal answer needs, yet fell short of a fully complete answer far more often. Check the response against every part of the original ask, review conclusions more closely as tasks get harder, and verify assertions on nuanced points of law against the source.
Are GPT-6.1 Sol and GPT-6.1 Terra available in Lito?
Yes. Both models are available in Lito now.
Is there Legal AI that works inside Microsoft Word?
Yes. Lito runs inside Word and Outlook, on the web, and on iOS, so lawyers draft, compare, and review where they already work, with no new platform to learn.
What is Lito?
Lito is Litera’s award-winning Legal AI agent for the practice and business of law. It combines leading foundational models with rules-based engines built over 30 years, skills curated by Litera in-house lawyers, and the firm’s own data.
These observations reflect a focused evaluation of each model’s default behavior in Lito by the Litera Legal Knowledge Engineering team at a point in time. Model behavior evolves, and individual results will vary with prompts and documents.
