APEX-Agents: Corporate Lawyer measures whether AI agents can execute real corporate lawyer tasks. Across 160 tasks graded by expert legal practitioners, the Fable 5 agent leads on the ReAct harness at 68.4% Mean Score, ahead of the Opus 4.8 agent (64.7%) and the GPT 5.4 agent (57.7%).
The table below shows Top 10 AI agent models for Corporate Lawyer tasks using Mean Scores for the ReAct harness. Scores are captured as of July 2026.
| Rank | Agent | Mean (ReAct) |
|---|---|---|
| 1 | Fable 5 | 68.4% |
| 2 | Opus 4.8 (Max) | 64.7% |
| 3 | GPT 5.4 (xHigh) | 57.7% |
| 4 | Opus 4.6 (Max) | 53.9% |
| 5 | GPT 5.5 (xHigh) | 53.6% |
| 6 | Opus 4.7 (Max) | 52.9% |
| 7 | Gemini 3 Flash (High) | 52.4% |
| 8 | Grok 4.5 | 52.2% |
| 9 | Gemini 3.1 Pro (High) | 50.5% |
| 10 | Gemini 3 Pro (High) | 48.7% |
The default leaderboard view uses the Loop harness; select ReAct to compare directly with these rankings.
View full APEX-Agents: Corporate Lawyer leaderboard →
How to read the APEX-Agents: Corporate Lawyer Agents leaderboard?
APEX-Agents evaluates each agent for corporate lawyer tasks on 2 harness architectures and scoring methodologies:
Harness Architectures:
- ReAct: the agent reasons step by step before acting, relevant for the Oil Pollution Act task where the agent must read the claim summary, apply the statutory authority, and reason through the survey in sequence.
- Loop: the agent operates in a truncated reasoning cycle, testing whether agents can integrate multiple legal sources without extended step-by-step planning.
Scoring:
- Pass@1: whether the agent completes the full task correctly on its first attempt. For the liability task, all 3 rubric criteria must be satisfied simultaneously.
- Mean Score: partial credit for each rubric criterion met. Identifying 1 party's liability correctly while missing the second and the equal-liability conclusion still earns credit for the criterion met.
Top 5 Corporate Lawyer AI Agent Models by ReAct, Pass@1:
1. Fable 5 40.9%
2. Opus 4.8 (Max) 37.5%
3. GPT 5.4 (xHigh) 29.8%
4. GPT 5.5 (xHigh) 29.3%
5. Opus 4.6 (Max) 26.5%
View full leaderboard (ReAct, Pass@1) →
Top 5 Corporate Lawyer AI Agent Models by Loop, Pass@1:
1. Fable 5 39.8%
2. GPT 5.6 Sol (Max) 35.6%
3. GPT 5.6 Sol (Max + Pro) 35.2%
4. Opus 4.8 (Max) 32.3%
5. GPT 5.6 Sol (xHigh) 32.3%
View full leaderboard (Loop, Pass@1) →
Top 5 Corporate Lawyer AI Agent Models by ReAct,Mean Score:
1. Fable 5 68.4%
2. Opus 4.8 (Max) 64.7%
3. GPT 5.4 (xHigh) 57.7%
4. Opus 4.6 (Max) 53.9%
5. GPT 5.5 (xHigh) 53.6%
View full leaderboard (ReAct, Mean Score) →
Top 5 Corporate Lawyer AI Agent Models by Loop, Mean Score:
1. Fable 5 67.4%
2. GPT 5.6 Sol (Max) 63.4%
3. GPT 5.6 Sol (Max + Pro) 63.1%
4. Opus 4.8 (Max) 60.3%
5. GPT 5.6 Sol (xHigh) 60.3%
View full leaderboard (Loop, Mean Score) →
Who should use APEX-Agents: Corporate Lawyer?
APEX-Agents: Corporate Lawyer serves 3 types of users:
- Individual corporate lawyers and associates evaluating which AI agent handles document-heavy legal analysis most reliably for their practice.
- Law firm technology and knowledge management teams standardizing AI agent deployments across practice groups. The benchmark gives them expert-graded agent performance data on tasks lawyers actually do.
- AI labs and legal tech companies building products for corporate law that need expert-graded performance data on real legal tasks.
How are AI agent rankings evaluated for corporate lawyer tasks?
These rankings come from APEX-Agents: Corporate Lawyer, Mercor's benchmark for how frontier AI agents perform on real legal work. Generic legal AI benchmarks test isolated research or drafting tasks. Corporate lawyer work requires agents to reason across multiple authoritative sources simultaneously and arrive at precise legal conclusions. APEX-Agents was built to score that.
What does the APEX-Agents: Corporate Lawyer benchmark measure?
APEX-Agents: Corporate Lawyer measures whether an AI agent running on frontier AI models can perform economically valuable corporate lawyer tasks across 160 cases. Tasks require agents to reason across multiple legal documents and authoritative sources simultaneously, using real legal tools. Each task is graded by a human-authored rubric with exact output requirements and scored as Pass@1.
Built with practitioners from Latham & Watkins, Skadden, and Cravath.
How does APEX-Agents: Corporate Lawyer benchmark scoring work?
Every rubric is written by an experienced corporate lawyer who defines what correct means for that task. Each criterion is a binary yes/no graded against a precise expected output. For the sample Oil Pollution Act task, one criterion reads: 'States that the maximum potential liability for Star Tankers International Ltd. under the Oil Pollution Act is $56,709,300.' In legal work, that figure is not a rounding issue. It is the conclusion that defines whether an agent's analysis would survive partner review.
What type of task is evaluated by the APEX-Agents: Corporate Lawyer benchmark?
The following sample task is provided for illustration only. Scores on this task may not reflect an agent's overall leaderboard performance.
Sample task: Oil Pollution Act liability analysis
The agent must evaluate maximum total potential liability for 2 parties, Star Tankers International Ltd. and Cooper/Jeffries Energy Corporation, under the Oil Pollution Act for a maritime incident involving the M/V Red Room. The agent must use 3 sources simultaneously: the BLPL Claim Summary, the relevant legal authority, and the Hull and Machinery Survey. It must calculate exact liability figures for each party, determine which party carries greater liability, and draft a memo communicating the conclusion.
The rubric has 3 criteria. On this sample task, the top-ranked agent (Fable 5 on ReAct) passed 1 of the 3: it correctly identified that the maximum potential liability for Star Tankers International Ltd. under the Oil Pollution Act is $56,709,300, but failed to correctly state the corresponding figure for Cooper/Jeffries Energy Corporation, and failed to reach the correct conclusion that both parties carry equal maximum liability under the Act.
What are the hardest corporate lawyer tasks for AI agents?
Multi-source statutory analysis and precise legal conclusion drafting remain the hardest tasks on APEX-Agents: Corporate Lawyer. The top-ranked agent scores 68.4% on ReAct Mean Score. On ReAct Pass@1, agents running GPT 4o score 3.4%, agents running o1 (High) score 3.2%, and agents running Gemini 2.5 Flash score 2.2%.
- Exact liability calculations are required. The rubric checks each figure against a precise expected value. A correctly reasoned answer with the wrong number still fails.
- Multi-source legal reasoning compounds errors. The sample task requires integrating 3 independent sources simultaneously. Agents that access each source individually often fail to apply them in combination.
- Legal conclusions must be reached, not just supported. Correctly identifying 1 party's liability while failing on the symmetric analysis of the second is still a failed task.
- Senior corporate lawyers still outperform AI agents on precision legal work. Multi-source statutory analysis and exact legal conclusion drafting remain largely the preserve of human expertise.
AI agents accelerate corporate legal work on research, drafting, and document review. They do not yet reliably produce final legal conclusions on complex multi-source analysis tasks without expert review.
Do AI models or AI tools for corporate lawyers matter more?
Both matter, but they play different roles. The underlying model like Fable 5, Opus 4.8, or GPT 5.4 determines the agent's reasoning capability and sets the performance ceiling. The tool like a legal research platform, a document management system, or a firm-specific AI assistant wraps that capability in context and delivers it into the lawyer's workflow.
A well-integrated legal tool cannot compensate for a weak underlying model on tasks requiring precise statutory analysis. Providing better document access does not fix an agent's failure to integrate 3 legal sources into a correct conclusion. However, a good tool amplifies what the underlying model offers, giving the agent structured access to case files, statutory databases, and precedent libraries.
For contract review and legal research summarization, the difference between the top models is smaller and your tool and workflow often decide the result.
For the precision-intensive work APEX-Agents measures, multi-source liability analysis and exact legal conclusion drafting, the underlying model's capability is the deciding factor. Pick the model for the ceiling you need; pick the tool for how effectively it delivers that capability into your practice.
Get the full APEX-Agents: Corporate Lawyer dataset
The public leaderboard shows the scores. The full dataset is what law firm technology teams and AI labs use to make real deployment decisions, including task-level breakdowns, rubric details, and agent trajectories across all 160 tasks.
For teams that need more
Evaluate AI agent performance on your own legal workflows with custom evaluations graded by Mercor's network of corporate lawyers.
Get in touchFrequently Asked Questions
What is the best AI agent for corporate lawyers right now?+−
As of July 2026, the Fable 5 agent leads APEX-Agents: Corporate Lawyer at 68.4% Mean Score on the ReAct harness, ahead of the Opus 4.8 agent (64.7%) and the GPT 5.4 agent (57.7%). On Pass@1, Fable 5 still leads at 40.9%, with Opus 4.8 at 37.5% and GPT 5.4 at 29.8%. Rankings shift with each new model release.
Will AI replace corporate lawyers?+−
The top-ranked agent scores 68.4% Mean Score on real corporate lawyer tasks on the ReAct harness. On Pass@1, the top agent completes only 40.9% of tasks on its first attempt, and agents running GPT 4o (3.4% Pass@1, 13.2% Mean Score) and o1 (3.2% Pass@1, 14.7% Mean Score) fail almost entirely on professional legal analysis tasks. AI agents accelerate legal work on research, drafting, and document review. They do not yet reliably perform the multi-source statutory analysis and precise legal conclusion drafting that corporate matters depend on.
How often is the APEX-Agents leaderboard for Corporate Lawyers updated?+−
The leaderboard is updated as new frontier models are evaluated. Rankings shift with each major release. Verify scores at the time of your decision on the live leaderboard.
How can my firm evaluate AI agents on our own legal workflows?+−
Mercor offers custom evaluations graded by vetted corporate lawyers on firm-specific legal tasks, giving law firm technology teams a benchmark built on their own matters. Get in touch.
Can I use ChatGPT or Claude on confidential client matters?+−
Enterprise tiers of most frontier model providers offer zero-data-retention options. Before using any AI agent on live client matter materials, confirm the specific data-handling terms for your tier, check your firm's information security policies, and review applicable professional responsibility rules on confidentiality in your jurisdiction.
How is the APEX-Agents: Corporate Lawyer benchmark scored?+−
APEX-Agents: Corporate Lawyer scores each task as Pass@1, measuring whether an agent fully completes a task on its first attempt. Each task has a human-authored rubric with binary yes/no criteria graded against exact expected outputs. Mean Score provides a complementary view, awarding partial credit for each criterion an agent gets right. The benchmark is evaluated across 2 harness architectures, ReAct and Loop, with ReAct Pass@1 as the primary ranking signal.
Can AI agents draft legal memos?+−
AI agents can draft legal memos, but producing a memo that would survive partner review on a precision legal question remains beyond current agents. The sample task on APEX-Agents: Corporate Lawyer requires the agent to draft a memo stating which party carries greater liability under the Oil Pollution Act. The top-ranked agent completes this task fully on only 40.9% of attempts on the ReAct harness, Pass@1. Output from AI agents on legal drafting tasks requires expert review before it reaches a client or partner.