Ivo releases Sage, an open legal model post-trained from DeepSeek V4 Flash for long-horizon contract agents

What changed

Legal-AI company Ivo announced Ivo Sage at its October 1 Inscribe event and published MIT-licensed weights on Hugging Face. Sage is built on DeepSeek V4 Flash and post-trained with reinforcement learning on long-horizon Legal Agent Benchmark tasks, targeting multi-tool-call work that produces memos, redlines or draft agreements from matter documents. Ivo's model card reports a criterion pass-rate increase from 70.1% for the base model to 91.3% on 55 held-out Contracts tasks; this is Ivo's own evaluation under a common protocol rather than an independent leaderboard result.