Microsoft optimizes AI efficiency with GPT-5.6 and internal token budgets

Microsoft is implementing strategic shifts in its AI deployment, focusing on the efficiency of the GPT-5.6 model family. Since August 2026, the company has prioritized GPT-5.6 within Excel Copilot to enhance multi-step data analysis. Benchmarks indicate that the GPT-5.6 Sol variant uses approximately 36% fewer tokens per Excel workbook compared to competitors like Opus 5, while resource consumption for creating PowerPoint decks has decreased by 21%.

Internally, Microsoft is also managing the costs of its AI integration by introducing token budgets for employees. Executive Vice President Jay Parikh stated in an internal email that the company aims to maximize results rather than “tokenmaxxing.” To optimize investment, Microsoft has made the more cost-effective GPT-5.6 the default internal setting. Each business unit now operates under specific monthly token budget targets, which employees can monitor via an internal dashboard.