Ivo releases open-weight Sage model for multi-step contract review
The AMW Read
Sage meaningfully updates Ivo's position through a deliberate open-weight strategy, but the demonstrated impact remains within contract AI and benchmark reliability is limited.
Named counterparties: Harvey
Ivo releases open-weight Sage model for multi-step contract review
San Francisco-based legal AI company Ivo has released Ivo Sage, a free, MIT-licensed contract model built on DeepSeek V4 Flash. According to LawFuel, Sage was post-trained with reinforcement learning on more than 1,000 tasks from the Legal Agent Benchmark. The reported results show it meeting 91% of individual contract criteria but completing only 16.4% of tasks perfectly. Reported cost per task was $1.24, compared with $142 for a leading frontier model. These figures describe benchmark performance, rather than demonstrated reliability across customers' production contracts.
The release brings an open-weight strategy into a legal AI market where Ivo competes for in-house counsel budgets alongside contract platforms and broader legal assistants. Its commercial products review agreements inside Microsoft Word against company negotiating playbooks and make signed contracts searchable. Ivo's argument is that context and workflow decomposition matter more than simply using a larger model: the company says it splits contract review into more than 400 separate AI tasks. Releasing Sage makes the model layer available to others while leaving a commercial question around the value of company-specific context, workflow integration, and contract intelligence. The gap between individual criteria and perfect task completion also cautions against treating partial benchmark success as dependable end-to-end execution.
For builders and investors, the concrete test is whether lower model costs translate into lower total review costs once verification and corrections are included. Sage gives teams a model they can inspect and build on, but its reported perfect-completion rate makes human review an important evaluation variable. Assessments should measure complete-task accuracy against a buyer's negotiating playbook, alongside inference cost and integration effort.
