Automating the 'Last Mile' of Finance

Model ML has developed an agentic workflow designed to handle the final, labor-intensive stages of financial analysis: reconciling evidence, formatting documents, and verifying data. By integrating GPT-5.6 Sol into their agent harness, the company enables finance professionals to generate editable, review-ready PowerPoint decks and Excel workbooks from briefs and raw source material. This transition has reduced the time required to build a bespoke tearsheet from one hour to approximately five minutes.

Performance Gains and Efficiency

Model ML’s internal benchmark, the 'Composite' evaluation, demonstrates that GPT-5.6 Sol outperforms previous models in both quality and efficiency:

• Excel Efficiency: The model uses 36% fewer tokens per workbook compared to Opus 5, while maintaining high headline accuracy.
• PowerPoint Readiness: GPT-5.6 Sol achieved a 43.3% professional-readiness rate (the ability to move directly to substantive review), compared to 26.7% for Opus 5. It also successfully produced a valid .pptx file in 100% of test cases.
• Agentic Integration: The system uses a core agent that plans tasks and routes specific sub-tasks to the most appropriate model, while providing the agent with toolkits for data integration, document editing, and code execution.

Shifting Toward Interactive Knowledge Work

Beyond static file generation, Model ML is moving toward 'surface-agnostic' and interactive outputs. By keeping documents connected to the underlying models and source data, reviewers can click through figures to inspect the financial models directly. The company argues that traditional Office software, designed for manual creation, is being fundamentally reshaped by AI, allowing for outputs that are dynamic rather than static.