Sol clears 43.3% of Model ML's finance gate

GPT-5.6 Sol produced an editable PowerPoint in 100% of Model ML’s finance-agent tests, yet only 43.3% cleared the company’s professional-readiness gate. The OpenAI case study and benchmark tables put Opus 5 at 76% and 26.7%, respectively, while Sol used 36% fewer tokens per Excel workbook but finished every key output correctly in 50% of cases versus Opus 5’s 60%. Finance teams should read the result through the archive’s task-specific model-routing framework: completion, review readiness, spreadsheet correctness, and token cost remain separate gates, so no single headline score can choose the production route.