Ivo releases Sage, an open model for long-horizon contract work
Ivo announced Sage, an open-source model post-trained for long-horizon contract work, at its first user conference, Ivo Inscribe. Ivo says attorneys, researchers and developers can download, run and build on the model for free.

Key takeaways · 4
- 01
Sage is free to download, run and build on, and can be fine-tuned on users’ own datasets.
- 02
Ivo says reinforcement learning raised Sage’s benchmark pass rate from 70% to 91%.
- 03
LawFuel reports Sage costs about US$1.24 per task, compared with US$142 for a leading frontier model.
- 04
Ivo’s preview benchmark found models often miss when to counter or reject a contract change, and add a needed new point only 23% of the time.
An open model built for contract work
Ivo announced Sage at Ivo Inscribe, its first user conference, describing it as the first open-source model from a legal AI company post-trained for long-horizon contract work.[1] Ivo says the model is free for attorneys, researchers and developers to download, run and build on, and that users can fine-tune it on their own datasets.[1] Its rules and safeguards can also be adjusted for specific use cases.[1] Ivo says Sage is available through Hugging Face and that it released the model with its research and evaluation record.[2]
What the performance figures measure
Ivo says it built Sage with River AI by post-training DeepSeek V4 Flash for long-horizon contract work; River AI provided technical and training-infrastructure support.[1] Ivo says training used public data and synthetic data generated by real attorneys.[1] On the Legal Agent Benchmark Contracts pass criteria, Ivo reports that Sage rose from 70% to 91% after reinforcement learning.[1] That should not be read as a 91% rate of perfect task completion: LawFuel reports that Sage completes 16.4% of contract tasks perfectly, a stricter measure.[3] LawFuel also reports a cost of about US$1.24 per task, against US$142 for a leading frontier model.[3]
A benchmark focused on review judgment
Ivo previewed a new contract-review benchmark developed with research partner micro1.[1] The benchmark evaluates issue prioritization, restraint, adapting to the deal, escalation and following a playbook.[1] Ivo says models get the right answer 82% of the time when they should accept a counterparty change or leave it unchanged, but 46% when they should counter or reject it.[1] Models add a needed new point of their own only 23% of the time.[1] Ivo also says models meet 51% of criteria for standard playbook positions on average, falling to 38% when the right move depends on deal facts.[1] A public set of benchmark results is expected in the coming weeks, according to Ivo.[1]
Deployment and evaluation caveats
LawFuel reports that Sage has 284 billion parameters and that Ivo’s setup instructions assume a multi-GPU server.[3] That makes the free model download distinct from a promise of cost-free or simple deployment: teams considering it should assess their own compute and integration requirements. Ivo says Sage achieved quality comparable to much larger frontier models with higher token efficiency at a fraction of the cost.[1] Those are Ivo’s claims, while the reported perfect-task completion rate and benchmark criteria pass rate describe different measures.[1][3] Separately, Ivo announced general availability of Collaborate, its contract-intelligence platform for Fortune 500 legal teams.[1]
Sage gives legal and technical teams an open model they can adapt and evaluate for contract work, while the benchmark figures underline that contract review involves judgment beyond matching standard positions. Teams should distinguish benchmark pass rates from perfect completion and account for the infrastructure needed to run the model.
Why it matters
Put this to work — one session a day, built for your industry.
Create a free account for a daily session — eight questions and one real-work challenge, on the news that affects your role.
Start freeHow this developed
7 October 2026
Ivo releases Sage, an open model for long-horizon contract work