Ivo has launched Sage, an open model tailored to contract work, built with River AI on DeepSeek V4 Flash. The pitch is that specialist training can make a smaller, cheaper model useful for legal teams, though the performance figures come from Ivo, not an independent evaluation.
Watch Desk analysis
What happened
Artificial Lawyer reports that Ivo and River AI post-trained DeepSeek V4 Flash on long-horizon contract work, using public material and synthetic data generated by lawyers. Ivo says the open model can be fine-tuned on law firms’ own data and adapted to their rules and safeguards. Read Artificial Lawyer’s report.
Ivo says the model’s score on the Legal Agent Benchmark’s Contracts pass criteria rose from 70% to 91% after reinforcement learning. Those are the company’s reported results; the article does not establish an independent comparison. Artificial Lawyer also cites substantially lower DeepSeek API input prices, but that is a comparison of model pricing, not a demonstrated cost for running Sage in legal workflows.
Why it matters
Legal AI vendors often sell tools built around general-purpose models. An open model aimed at contract work gives law firms another possible route: adapt the system to their documents and playbooks, rather than relying solely on a vendor’s fixed model. That flexibility may appeal to teams with particular workflows, but it also means they need the people and infrastructure to test and maintain their own versions.
The 70%-to-91% result is a promising claim, not a verdict on how Sage performs across real contracts, jurisdictions or firms. Benchmark scores are useful starting points; legal work has a habit of asking awkward questions beyond the test set.
Our read
The interesting news is the combination of a legal specialism and an open model that firms can adapt. That is more consequential than a price comparison assembled from API rates. Legal teams should look for the benchmark methodology, licensing terms and results on their own documents before treating the launch as a ready-made alternative to larger models.
What to watch
- Whether Ivo publishes more detail on the benchmark setup and evaluation results.
- What the model’s licence permits, and what firms need to run or fine-tune it.
- Whether independent or customer testing supports Ivo’s accuracy and cost claims.
Discussion spark: Would you trust a fine-tuned open model with contract work if it scored well on a benchmark, or should legal teams require independent testing on their own documents first?
Sources and evidence
- Ivo Launches Open-Source DeepSeek Contract AI Model – Artificial Lawyer (7 October 2026, 08:51 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.