Microsoft says Copilot costs less than rivals in a study it funded and ran
The comparison includes usage charges, not just subscriptions. Its estimates depend on selected workloads and pricing assumptions, including unpublished prices.
Loading page…
The comparison includes usage charges, not just subscriptions. Its estimates depend on selected workloads and pricing assumptions, including unpublished prices.
Listen to this story
Microsoft’s cost comparison models what two work patterns would cost across Microsoft 365 Copilot, ChatGPT Enterprise and Claude Enterprise, including subscriptions and usage charges. Its estimates favor Copilot, while reported task quality was broadly similar across the tested workloads. Because Microsoft funded and conducted the testing and cost analysis, with GigaOm helping design and validate the work, the results are directional rather than independent customer pricing. Buyers should rerun the evaluation using their own tasks, quality standards and negotiated prices.
The October 2026 evaluation tested 37 prompts across Copilot Chat and Cowork, plus Copilot in Word, Excel and PowerPoint.
The monthly scenarios modeled 100 tasks for everyday users and 140 for advanced users, including 20 agent workflows.
Copilot’s everyday chat and app use carried no additional usage charges in the study; advanced agent workflows were billed through Copilot Credits.
Microsoft published research on November 16 putting Microsoft 365 Copilot’s estimated monthly cost below ChatGPT Enterprise and Claude Enterprise for two modeled types of worker. The study found broadly similar overall task quality. But the comparison comes with a distinction buyers should know upfront: Microsoft funded the research and performed the testing and cost analysis itself.
GigaOm helped design the research, validated its methodology, results and cost analysis, and co-developed the findings. The study used hands-on testing conducted in October 2026. Microsoft’s announcement presents the results as a comparison of what completing the selected work costs, rather than a comparison of subscription prices alone.
The largest difference came from everyday work. Copilot’s subscription includes chat and Microsoft 365 app usage, so those tasks incurred no additional usage charges in the study. The competing products’ modeled bills combined subscription and usage costs. Copilot’s more advanced agent workflows—multi-step jobs that run on their own—were billed separately through Copilot Credits.
Finding workplace information was another factor in Microsoft’s explanation. ChatGPT and Claude accessed Microsoft 365 data through connectors and add-ins. Microsoft says Copilot’s Work IQ technology uses an organized index of workplace content to locate relevant information, instead of relying solely on keyword searches or repeatedly processing whole files. That is designed to reduce how much information a model must process.
The third factor is model selection. Microsoft says Copilot’s Auto routing chooses models and effort levels for different parts of a job. Its stated aim is to spend more computing power where that improves the result, rather than automatically using the most powerful model at its highest effort for every step.
Testing covered 37 unique prompts across Copilot Chat and Cowork, plus Copilot in Word, Excel and PowerPoint. Each product worked against the same fictional company and underlying data. Prompts were run repeatedly to account for variation in AI responses, with monthly costs modeled around two patterns of use:
Before testing, human evaluators wrote a checklist for each prompt. It required the correct sources, people, facts and figures, a complete deliverable, and no unrequested actions. A response passed only if it met every requirement; partially correct work failed. The broadly similar quality finding applies across the workload, not to every individual task, where results varied.
The estimates use October 2026 list prices, or assumed prices where published prices were unavailable. They exclude negotiated discounts. Microsoft describes the findings as directional because task selection, frequency, product settings and customer pricing can change the outcome. These figures are therefore workload estimates, not universal monthly prices for the three products.
For buyers, GigaOm recommends repeating the exercise with representative tasks, multiple runs, defined quality checks and their own pricing. Microsoft says its account teams can share the evaluation methodology and full prompt list, giving customers a starting point for testing the cost claim against their own work.
Loading discussion...
Join the conversation
Explain what would make the findings convincing—or leave you skeptical.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.