Paper Instruments said Ultramarine 1 Flash was post-trained by a team of five for roughly $300k all-in, and that it is available on API now, with weights to follow within two weeks. The company described the release as one of the largest open experiments in non-verifiable agentic RL, with training code and a tech report open-sourced. The model is positioned as a drop-in alternative to GPT and Claude for AI companies serving banks, law firms and other knowledge-work clients. The author said prolonged training taught professional standards such as visual hierarchy and straightforward prose.

AI summary of the linked source β€” not editorial content.

Loading related coverage…

Comments

Loading comments…