GPT-5.6 Sol completes finance decks in 100% of tests

Model ML's evaluation is notable for focusing on the 'last mile' of finance work—the point where analysis must be turned into editable, traceable, and review-ready files. The startup's Composite benchmark tests not just whether a model can produce a plausible document, but whether that document holds up to scrutiny, with correct numbers, traceable sources, and editable charts. GPT-5.6 Sol's high completion rate on PowerPoint generation, paired with a higher pass rate on the 'professional-readiness' gate than Opus 5, positions it as a strong tool for this specific, high-stakes use case.

The context of this comparison is a shift in how finance professionals work. Model ML's customers are moving toward browser-based outputs that stay connected to the underlying models and source material. The company's co-founder argues that PowerPoint, Excel, and Word were 'designed for a world where creating knowledge work was manual,' and that 'AI has changed that assumption.' This suggests document creation is becoming less about manual formatting and more about the agent's ability to plan, reason, and maintain context over a long workflow.

Whether GPT-5.6 Sol becomes a default choice in finance likely hinges on how its results generalize beyond Model ML's specific harness. The benchmark shows efficiency gains—for example, using about 21% fewer tokens than Fable 5 on PowerPoint—but also some trade-offs, such as slightly lower visual quality scores than Opus 5 on specific sub-metrics. For Model ML customers, the immediate impact is tangible: deliverables that took an hour can now take minutes, shifting the professional's role from rebuilding analysis to refining judgment.