Ivo launched Sage, an open-source model for contract work, on October 7, 2026. Built with River AI, Sage was created by post-training DeepSeek V4 Flash on long-horizon contract work. Its training used public data and synthetic data generated by lawyers, while River provided technical and training infrastructure. Ivo, founded in New Zealand and now also operating from San Francisco, says lawyers can fine-tune the model on their own datasets, adjust its rules and safeguards for particular uses, and test new approaches.

After reinforcement learning, Sage’s score on the Legal Agent Benchmark (LAB) Contracts rose from 70% to 91% of the benchmark’s pass criteria. Ivo said the model reached quality comparable to much larger frontier models while using tokens more efficiently and costing a fraction as much. The article reports DeepSeek V4.1 Flash API pricing at $0.15 per million uncached input tokens, compared with $2 to $5 per million input tokens and $12 to $30 per million output tokens for a comparable frontier model; ChatGPT’s comparison characterized the difference as around 20X to 50X cheaper.

Ivo co-founder and CEO Min-Kyu Jung said the next advance in contract and legal AI depends on giving models the work context, including documents, playbooks and how a legal team operates, rather than simply making models larger. He said Ivo is opening Sage so others can test, adapt and build on it. The article says lawyers can already select from multiple models on many legal AI platforms, and identifies Sage’s open-source availability and lower cost as distinguishing features; it adds that the price could undercut competitors’ economics even when their platforms route requests to cheaper model combinations.

Ivo launched Sage, an open-source model for contract work, on October 7, 2026. Built with River AI, Sage was created by post-training DeepSeek V4 Flash on long-horizon contract work. Its training used public data and synthetic data generated by lawyers, while River provided technical and training infrastructure. Ivo, founded in New Zealand and now also operating from San Francisco, says lawyers can fine-tune the model on their own datasets, adjust its rules and safeguards for particular uses, and test new approaches.

After reinforcement learning, Sage’s score on the Legal Agent Benchmark (LAB) Contracts rose from 70% to 91% of the benchmark’s pass criteria. Ivo said the model reached quality comparable to much larger frontier models while using tokens more efficiently and costing a fraction as much. The article reports DeepSeek V4.1 Flash API pricing at $0.15 per million uncached input tokens, compared with $2 to $5 per million input tokens and $12 to $30 per million output tokens for a comparable frontier model; ChatGPT’s comparison characterized the difference as around 20X to 50X cheaper.

Ivo co-founder and CEO Min-Kyu Jung said the next advance in contract and legal AI depends on giving models the work context, including documents, playbooks and how a legal team operates, rather than simply making models larger. He said Ivo is opening Sage so others can test, adapt and build on it. The article says lawyers can already select from multiple models on many legal AI platforms, and identifies Sage’s open-source availability and lower cost as distinguishing features; it adds that the price could undercut competitors’ economics even when their platforms route requests to cheaper model combinations.