Paper Instruments says Ultramarine 1 Flash does office work for $0.13 a task

Paper Instruments has released details of Ultramarine 1 Flash, a model for professional knowledge work such as diligence memos and financial analysis. It is built by applying large-scale reinforcement learning to the open GLM-5.3 Flash model, training on tasks that cannot be checked by simple tests, with grading rubrics written per task. The lab says it beats GLM-5.3 Flash on all six knowledge-work benchmarks it ran, by 4.1 points on average, while using 6.5% fewer tokens. On the lab's own internal knowledge-work suite, Paper Instruments says Ultramarine costs $0.13 per task, outscores Claude Fable 5 ($8.19) and GPT-5.6 Sol ($2.09), and comes within three points of Claude Opus 5.5 ($6.96) and GPT-6 Sol ($1.26), with all four rivals at maximum reasoning. These are the lab's own results. The post does not say how to access the model, and there is no independent test of these claims yet.