Financial work doesn’t end when the data appears. Next come the formulas, sources, charts, formatting, and a careful review to make sure everything can be defended before a client or investment committee. Model ML says GPT-5.6 Sol helps automate much of that final stretch.
From a request to a file ready for review
Model ML develops AI agents for financial teams. These agents can receive a request, research information, perform calculations, and deliver an editable PowerPoint or Excel file.
The idea isn’t for AI to replace professional judgment. Its role is to handle repetitive and technical tasks, such as gathering evidence, applying formulas, organizing multiple spreadsheets, or preparing a presentation with traceable sources.
The user can focus on adjusting assumptions, reviewing the data, and improving the message instead of rebuilding the entire analysis from scratch.
The system can also begin a task from an email, continue it in the Model ML application, and finish it through Microsoft Office add-ins, without requiring the user to explain the context again.
Less time and fewer resources to create reports
According to Model ML’s evaluations, GPT-5.6 Sol showed relevant improvements over other models in financial workflows:
- It used 21% fewer tokens per PowerPoint presentation than Fable 5.
- It used 36% fewer tokens per Excel workbook than Opus 5.
- It achieved a 16.6 percentage-point advantage in professional readiness over Opus 5.
- It reduced the creation of a customized financial report in one use case from approximately one hour to five minutes.
In another workflow, the agents processed virtual data rooms containing more than 100,000 rows and hundreds of files in a single operation. This could be especially useful for mergers and acquisitions, investment analysis, and document review processes.
Of course, reducing the time doesn’t eliminate the need for supervision. In finance, one incorrect number or a poorly connected source can completely change a decision.
Editable presentations, not simple images
One of the biggest challenges for AI working with documents is producing files that are truly usable. A presentation can look attractive and still fail if its charts have been converted into images, its figures lack support, or its slides need to be rebuilt before they can be shared.
Model ML says GPT-5.6 Sol completed its PowerPoint workflow in 100% of the evaluated cases, compared with 76% for Opus 5. It also reached the level of professional readiness in 43.3% of the tests, compared with 26.7% for Opus 5.
The system works with tools that can create charts, tables, and editable elements. It also preserves the original request throughout the process and visually reviews each slide before delivering the file.
In other words, this isn’t just about asking AI to write a summary. The goal is to obtain a document that an analyst can open, review, modify, and present.
Excel with formulas, logic, and traceability
For Excel tasks, the agent can start with a client template or an empty workbook. It then gathers the necessary data, builds formulas across different tabs, and applies common financial analysis formatting.
The goal is to produce a model that can be recalculated and audited. For a professional, this means being able to review how a figure was reached, change an assumption, and see the effect across the rest of the file.
Traceability is key. When a piece of data comes from a document, database, or external source, the team needs to know where it came from and how it was used. That ability can make the difference between an interesting draft and material ready for review.
The importance of the agent and its tools
GPT-5.6 Sol doesn’t work in isolation. Model ML integrates it into an agent system that decides how to organize the task, which tools to use, and which model is best for each step.
The environment can load tools to access data, edit documents, and run code. This structure allows the agent to focus on the objective and use only the resources needed to complete the assignment.
Model ML explains that it arrived at this configuration through working sessions with OpenAI. During these sessions, they analyzed how the agent planned presentations, selected tools, and maintained context. They then adjusted its instructions and defined when each set of tools should be activated.
Does this mean that any financial report will be perfect with one click? No. AI can speed up preparation, but reviewing assumptions, sources, and conclusions remains the team’s responsibility.
The future points toward connected documents
Model ML also sees a transition from static files to results that remain connected to the original models and sources.
In that scenario, a professional could open an investment summary in the browser, select a figure, consult the financial model supporting it, and chat with the agent from the same page. The document would stop being an endpoint and become a work interface.
The proposal is easy to understand: less time building documents and more time making decisions. For that promise to be reliable, AI must deliver results that are editable, verifiable, and easy to review.
GPT-5.6 Sol represents a step in that direction within Model ML’s financial workflows. Its real impact won’t depend only on producing files faster, but on whether those files can withstand the difficult questions that arise in any professional review.