Ultramarine 1 Flash Model Targets Professional Knowledge Work Efficiency
New AI model delivers Fable-class performance for finance and operations at significantly lower costs.
1Professional Knowledge Work Focus
Ultramarine 1 Flash is specifically engineered for high-stakes professional environments including finance, law, consulting, and operations. Unlike general-purpose models, it is optimized to produce finished Excel workbooks, PowerPoint decks, and complex Word documents. It also integrates directly into live business systems such as ERP and MES to execute professional-grade tasks.
2Benchmarking Performance Metrics
On the JobBench evaluation, Ultramarine 1 Flash achieved a score of 62.9, outperforming Gemini 3.8 Flash at 60.6 and DeepSeek V4.1 Flash at 60.4. The model also demonstrated strong capabilities in the PI Knowledge Work internal benchmark, scoring 59.3, which places it ahead of Claude Fable 5 and GPT-5.6 Sol models.
3Cost-Efficiency and Token Usage
The model offers a significant economic advantage, costing only $0.13 per task. This price point is roughly 60 times cheaper than Claude Fable 5, which costs $8.19 per task. Additionally, Ultramarine 1 Flash improves token efficiency by 6.5% compared to its base model, GLM-5.3 Flash, by taking a shorter path to results.
4Rigorous Evaluation Methodology
Performance is measured using the PI Knowledge Work benchmark, which correlates strongly with the Artificial Intelligence Index at r = 0.81. Each task is assessed via a specific rubric focusing on substance, accuracy, and presentation. All models are tested within the same harness using official API pricing to ensure fair, transparent cost-quality comparisons.
5Professional Output Standards
Ultramarine 1 Flash is designed to adhere to the strict stylistic and structural rules required of analysts in professional services. It generates clean, straightforward prose while maintaining a sophisticated sense of visual and information hierarchy. This ensures that outputs are ready for immediate use in corporate reporting and high-level business presentations.
6Data Reconciliation Capabilities
The model demonstrates advanced reasoning in complex data tasks, such as reconciling five periods of segment sales and earnings in a single table. It successfully handles calculations where subtotals and totals must align perfectly. For instance, it accurately tracks financial shifts, such as steel mill earnings falling from $7.2bn to $2.4bn.
7Competitive Market Positioning
Ultramarine 1 Flash occupies a unique position on the cost-quality frontier. While it remains highly affordable at $0.13 per task, it performs within three points of premium models like Claude Opus 5.5 and GPT-6 Sol. This makes it a viable alternative for firms seeking high reasoning capabilities without the premium pricing.
8Strategic Implications for Enterprises
By combining specialized professional training with lower operational costs, Ultramarine 1 Flash addresses the need for scalable AI in enterprise workflows. As professional services firms increasingly automate document creation and data reconciliation, the ability to maintain high accuracy at a fraction of the cost becomes a critical factor for long-term operational strategy.