Model ML Expands GPT-5.6 Sol After 100% Deck Completion in Finance Tests
Updated
Updated · OpenAI · Aug 10
Model ML Expands GPT-5.6 Sol After 100% Deck Completion in Finance Tests
3 articles · Updated · OpenAI · Aug 10
Summary
Model ML widened GPT-5.6 Sol’s production role after internal Composite evaluations showed it produced more review-ready finance decks and spreadsheets for client-facing work.
100% of GPT-5.6 Sol PowerPoint runs finished successfully versus 76% for Opus 5, while its professional-readiness rate reached 43.3% against 26.7%, giving Model ML evidence to shift workflows from Opus 4.8.
36% fewer tokens per workbook than Opus 5 in Excel and about 21% fewer than Fable 5 in PowerPoint helped the company cut cost while keeping outputs editable, traceable and ready for substantive review.
At one global asset manager, a tearsheet that took about 1 hour now takes roughly 5 minutes, and Model ML says its agents can also process data rooms with more than 100,000 rows and hundreds of files in one pass.
Model ML says the push reflects a broader shift in finance software toward browser-based, source-linked outputs where users review assumptions and messaging instead of rebuilding decks and models by hand.
How does GPT-5.6 Sol guarantee data privacy when processing massive virtual data rooms for highly sensitive financial deals?
If AI agents perfectly format financial decks, will human analysts gradually lose their deep contextual understanding of the underlying data?
What hidden risks emerge when highly polished, AI-generated financial models contain deeply embedded calculation errors that bypass standard human review?