The context · January 2025
DeepSeek released R1 in January 2025, bringing another openly available reasoning model into the conversation about how organisations build and operate AI. [1]
A proposed project brief inspired by the period’s developments. This is an idea for exploration, not an announced Sansa product, an approved roadmap or a claim of completed client work.
A bounded comparison
Choose an internal knowledge task with an identifiable owner and a modest set of approved documents. Create representative questions, including ambiguous requests and questions the documents cannot answer. Record what a correct answer must contain before running either model.
Measure the cost of a finished answer
Run the same questions against a hosted service and a self-managed option. Have reviewers assess the answers without seeing which system produced them. Record review effort, retrieval failures and operational work alongside compute costs. Use synthetic or approved data until the handling arrangements are agreed.
What the project would deliver
The output would be a comparison report, a reusable evaluation set and a deployment recommendation with its assumptions. Proceed only if the chosen approach meets an agreed quality floor and has a named operating owner. A recommendation to stay with the existing service is a valid result; this concept does not assume a migration.
Source & context
DeepSeek · R1 release, 20 January 2025This retrospective was written for the archive in September 2026. The linked primary source documents the announcement or event; the practical interpretation and proposed approach are Sansa’s editorial perspective. Public examples do not imply a client relationship. Product capabilities and guidance may have changed since the period discussed.
Another perspective · January 2025
After R1: what does model independence actually buy?
Continue readingWorking through a similar question?
Talk it through with Sansa