Model ML, founded by brothers Arnie and Chaz Englander after two exits and experience running a private family office, builds AI agents that automate the final mile of financial analysis: reconciling evidence, building editable PowerPoint and Excel files, and linking every claim to its source. A core agent plans the work, selects tools, runs calculations, and routes each step to the best model—often GPT‑5.6 Sol. The agent uses Model ML’s own document creation tooling to produce native Office files with traceable sources. The platform is “surface-agnostic,” letting users start in email or the Model ML app and continue via Microsoft Office plug-ins without re-explaining context.

In the company’s Composite evaluation benchmark for financial services, GPT‑5.6 Sol outperformed leading models on token efficiency and completion rates. For PowerPoint workflows, GPT‑5.6 Sol completed 100% of test cases (vs. 76% for Opus 5) and passed Model ML’s professional-readiness gate—output ready for substantive review—in 43.3% of cases (vs. 26.7% for Opus 5). It also used about 21% fewer tokens than Fable 5 while producing competitive deck quality. For Excel workflows, GPT‑5.6 Sol used 36% fewer tokens per workbook than Opus 5. Model ML has expanded GPT‑5.6 Sol in production, replacing Opus 4.8 in some workflows.

At a global asset manager, a bespoke tearsheet that took an analyst an hour now takes about five minutes. Model ML agents have processed virtual data rooms with over 100,000 rows and hundreds of files in a single pass. The agent’s harness provides toolkits for data integrations, document editing, and code execution, refined through on-site sessions with OpenAI. Customers are moving toward browser-based outputs that stay connected to live models and sources. As Chaz Englander puts it, “PowerPoint, Excel, and Word were designed for a world where creating knowledge work was manual. AI has changed that assumption. The software itself is about to change.”