Model ML completes finance work more efficiently with GPT-5.6 Sol
Model ML integrates GPT‑5.6 Sol into finance workflows, reducing token usage by 21% per PowerPoint deck and improving Excel and presentation outputs compared to prior models.
Model ML, a software platform built by Arnie and Chaz Englander after two exits, automates finance workflows from research to editable deliverables. Its agents use GPT‑5.6 Sol as a core model to plan tasks, reconcile evidence, and generate native PowerPoint and Excel files with traceable sources. The system supports continuity across email, the Model ML app, and Microsoft Office plug-ins, allowing finance teams to start and finish assignments without re-explaining context.
In Model ML’s Composite benchmark, GPT‑5.6 Sol completed 100% of PowerPoint workflows versus 76% for Opus 5, and met the professional-readiness threshold in 43.3% of cases compared to 26.7%. For Excel, it used 36% fewer tokens per workbook than Opus 5. Real-world use cases show time savings, such as a tearsheet reduced from one hour to five minutes and processing of virtual data rooms with over 100,000 rows in a single pass.
Model ML’s benchmarks evaluate assignments from initial briefs through research, calculations, and editable outputs, scoring factors like number accuracy, source traceability, formula correctness, and presentation quality. GPT‑5.6 Sol outperformed Opus 5 in deck quality, brief adherence, hierarchy, and consistency. The company now uses GPT‑5.6 Sol in production for some workflows previously handled by Opus 4.8, citing improved reliability and reduced token usage.
Model ML’s platform generates secure, interactive outputs that remain connected to underlying models and source material. The company plans to expand browser-based workflows where reviewers can access financial models and interact with agents directly from the output. Englander notes that traditional office software was designed for manual creation, and AI is changing that foundation.