Model ML and GPT-5.6 Sol for Finance Workflow Automation

GPT-5.6 Sol Automates the "Last Mile" of Financial Analysis

Model ML has integrated GPT-5.6 Sol to automate the final stages of financial analysis, enabling the creation of review-ready, editable PowerPoint decks and Excel workbooks from initial briefs and source materials. This integration addresses the "last mile" of finance work—reconciling evidence, formatting files, and linking claims to sources—reducing the time required for specific deliverables, such as bespoke tearsheets, from approximately one hour to five minutes.

End-to-End Finance Workflow Capabilities

Model ML employs a "surface-agnostic" approach, allowing finance professionals to initiate assignments via email or the Model ML app and continue them through Microsoft Office plug-ins. The system utilizes a core agent that plans work, selects tools, and routes tasks to the most suitable model, frequently GPT-5.6 Sol.

Key capabilities include:

Native PowerPoint Creation: The agent transforms briefs and source materials into editable PowerPoint presentations with traceable sources.
Native Excel Creation: The agent can utilize client templates or blank workbooks to gather data, build formulas and logic across multiple tabs, and apply finance-specific formatting.
Large-Scale Data Processing: Model ML agents have processed virtual data rooms containing hundreds of files and over 100,000 rows of data in a single pass.

Performance Benchmarks: GPT-5.6 Sol vs. Competitors

Using its "Composite" evaluation benchmark, Model ML tested GPT-5.6 Sol against other models including Opus 5, Fable 5, and GPT-5.5 across PowerPoint and Excel workflows.

PowerPoint Workflow Performance

GPT-5.6 Sol demonstrated superior deliverability and completion rates:

Completion Rate: GPT-5.6 Sol completed 100% of test cases, compared to 76% for Opus 5.
Professional-Readiness Rate: 43.3% of GPT-5.6 Sol's outputs cleared the professional-readiness gate, a significant increase over Opus 5 (26.7%) and Fable 5 (32.0%).
Quality Metrics: GPT-5.6 Sol led Opus 5 in deck quality, brief adherence, hierarchy, and consistency.
Efficiency: GPT-5.6 Sol used approximately 21% fewer tokens than Fable 5 for PowerPoint workflows.

Excel Workflow Performance

In Excel workflows, GPT-5.6 Sol showed marked efficiency gains:

Token Efficiency: GPT-5.6 Sol used 36% fewer tokens per workbook than Opus 5 (2.44M vs 3.83M tokens).
Accuracy: It achieved a headline accuracy of 83.3%, matching GPT-5.5 and slightly exceeding Opus 5 (82.8%).
Speed: The model averaged 7.0 minutes per workbook, slightly faster than Opus 5 (7.5 minutes).

Technical Implementation and Agent Harness

Model ML developed its agent harness through on-site sessions with OpenAI to refine how the agent plans presentations, selects tools, and maintains context. The harness provides the agent with specific toolkits for data integrations, document editing, and code execution environments, ensuring the agent remains focused on the necessary tasks.

To ensure the output is "ready for real work," the agent maintains the original brief in context and performs a visual review of every slide before returning the final editable file.

Future of Financial Knowledge Work

Model ML is transitioning toward browser-based, interactive outputs that remain connected to the underlying models and source material. This allows reviewers to click through from an investment summary directly to the supporting financial model, moving away from the manual nature of traditional software like PowerPoint and Excel.