We were joined by Senne Mennes, co-founder of Clausebase, who explored the practical implications of OpenAI’s latest language model for legal practice.
The Reality of GPT-5: Evolution, Not Revolution
Mennes characterised GPT-5 as “an evolution, not a revolution,” noting that whilst improvements exist, the model falls short of the transformative leap many anticipated following GPT-4’s release. The primary enhancement centres on instruction-following capability, which proves critical for legal applications where AI tools must execute complex, layered instructions accurately—from proofreading and clause extraction to document automation.
Layout Awareness and Document Styling
A significant challenge in legal technology involves document styling. Mennes explained that legal documents typically suffer from “extremely broken” styling, with lawyers lacking proper knowledge of Microsoft Word’s features. This creates a complex technical challenge: AI systems must interpret both official styling rules and how lawyers have actually implemented styles in practice.
GPT-5’s enhanced instruction-following capability directly improves this process, enabling more reliable style interpretation during document modification. Notably, even Microsoft’s own Copilot cannot maintain document styling when rewriting content, typically returning plain text versions.
Advanced Features and Market Adoption
The new model introduces orchestration that automatically selects between computational models based on task requirements, eliminating the need for users to manually determine which variant to use. This proves particularly valuable for information extraction and complex analytical work.
Despite technological advances, legal sector adoption remains inconsistent. Mennes observed that adoption depends more on individual personalities than organisational size, though larger firms possess greater budgets for exploration. Many organisations lack understanding of fundamental concepts like context windows, necessitating significant education alongside demonstrations.
Interestingly, AI awareness has prompted renewed interest in established technologies. Mennes described situations where firms seeking AI solutions actually need traditional document automation—technology available for thirty years but only now receiving attention.
Language Support and Commercial Implications
Preliminary testing suggests GPT-5 maintains or improves performance across multiple languages, with Mennes reporting that Dutch-language performance matches English quality. However, costs will likely increase as orchestration automatically engages more expensive computational models.
Business Model Impact
Despite technological capabilities, billable hour structures remain largely unchanged. Hourly rates continue increasing, and whilst alternative fee arrangements receive more discussion, fundamental business model transformation has not materialised. Some clients have implemented productised services using document automation, but these remain exceptions.
Future Outlook
The speakers expressed uncertainty about GPT-6’s timeline and impact. Mennes suggested that OpenAI may not maintain market leadership, with the competitive landscape becoming increasingly balanced as multiple providers offer specialised capabilities.
Conclusion
GPT-5 represents meaningful progress in instruction-following for legal applications, particularly benefiting document styling and analytical tasks. However, the legal profession’s primary challenge remains systematic adoption of existing capabilities rather than technological limitations. The emphasis should be on practical engagement with available tools through pilot projects rather than waiting for revolutionary advances.
