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 complex professional domains including finance, law, consulting, and operations. Unlike general-purpose models, it excels at producing finished Excel workbooks, PowerPoint decks, and Word documents. It is also designed to operate directly within live business systems like ERP and MES, mirroring the daily workflows of analysts.
2Benchmark Performance Metrics
The model achieves a score of 62.9 on the JobBench evaluation, outperforming Gemini 3.8 Flash at 60.6 and DeepSeek V4.1 Flash at 60.4. In the PI Knowledge Work internal benchmark, it reaches 59.3, surpassing Claude Fable 5 at 57.8 and GPT-5.6 Sol at 55.1, demonstrating strong capabilities in high-stakes business reasoning tasks.
3Cost-Quality Frontier Advantage
At $0.13 per task, Ultramarine 1 Flash offers a highly competitive price point compared to industry alternatives. It significantly undercuts Claude Fable 5, which costs $8.19 per task, and GPT-5.6 Sol at $2.09 per task. This positioning allows professional firms to maintain high output quality while drastically reducing their operational AI expenditure.
4Token Efficiency Improvements
Across six distinct benchmarks, Ultramarine 1 Flash generates 6.5% fewer tokens than its base model, GLM-5.3 Flash. This reduction indicates a more direct path to achieving high-quality results. By minimizing unnecessary token generation, the model provides faster execution times and lower API costs per million tokens, matching the pricing of GLM-5.3 Flash.
5Rigorous 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 evaluated against a specific rubric focusing on substance, accuracy, and presentation. All models are tested in the same harness using official API pricing to ensure a fair, transparent comparison.
6Professional Output Standards
The model is built to adhere to the strict formatting and structural rules required of analysts in professional services. It produces clean, straightforward prose while maintaining a sophisticated sense of visual and information hierarchy. This ensures that generated outputs, such as financial tables, remain accurate, reconciled, and ready for immediate client presentation.
7Data Reconciliation Capabilities
Ultramarine 1 Flash demonstrates precision in handling complex data sets, such as multi-period segment sales and earnings tables. It ensures that every subtotal and total reconciles perfectly within the output. For example, it can accurately track shifts in steel mill earnings, which fell from $7.2bn in 2022 to $2.4bn in 2025.
8Strategic Market Positioning
By bridging the gap between high-reasoning capabilities and low-cost deployment, Ultramarine 1 Flash challenges established models like Claude Opus 5.5 and GPT-6 Sol. Its ability to perform at a fraction of the cost makes it a viable tool for scaling automated professional services, effectively shifting the economics of AI-driven business operations.