Copilot Studio: Licensing & Credits 2026

Key insights

• Copilot Credits: The universal usage unit for Copilot Studio.
  Credits pay for agent actions like summarizing, generating content, or calling tools.
  Consumption depends on three factors: the harness you use, the depth of interaction, and which features or models the agent calls.
• How to buy: Four purchase paths give different price and flexibility trade-offs.
  Options are Pay-As-You-Go (metered at about $0.01 per credit), a 25,000-credit capacity pack (~$200/month), Copilot Credit Pre-Purchase Plans (P3), and the Microsoft Agent P3 pool of Agent Commit Units (ACUs).
  Watch for the capacity pack carry-forward rules before you commit.
• Harness types: Copilot Studio uses three harnesses with different billing profiles.
  The GitHub Copilot harness can start consuming credits while you author (light/medium/heavy authoring tiers). The Standard harness and Copilot Chat harness bill per answers/actions at runtime.
  Note: Microsoft 365 Copilot covers Standard and Copilot Chat usage but does not cover GitHub Copilot harness consumption.
• Cost drivers and runtime tiers: Runtime consumption maps to light, medium, and heavy scenarios.
  Key drivers are model choice, runtime duration, context size, and use of external tools. Expect different credit rates for classic answers, generative answers, agent actions, and tool calls.
• Pricing math and units: Understand the conversion and tier discounts.
  Copilot Credit Commit Units (CCCU) equate $1 to 100 credits and offer tiered discounts up to ~20%. Agent Commit Unit (ACU) pools are available at larger commitment levels (examples include 20k, 100k, 500k ACU tiers).
• Design and procurement tips: Save credits by moving logic out of LLM calls.
  Manual, non-LLM configuration consumes zero credits, so precompute, cache, or handle simple rules without the model.
  Start with metered or small capacity packs to measure real usage before buying large precommit plans, and re-check Microsoft’s official licensing guide regularly because terms change.

Video overview and context

In a recent YouTube video, Dhruvin Shah [MVP] explains Microsoft’s August 2026 rewrite of Copilot Studio licensing and shows why the change matters for anyone building agents. The video focuses on the shift to usage-based billing through Copilot Credits, and it also walks through how different harnesses and purchase plans affect costs. Consequently, this is essential viewing for architects and procurement teams who must forecast ongoing expenses rather than only one-time setup costs.

Moreover, Shah provides demos and a chaptered breakdown that clarify when billing begins and how authoring activities can immediately consume credits. He emphasizes that Microsoft updates licensing frequently and recommends verifying the latest official guidance before committing to any plan. Therefore, readers should treat the video as a practical guide rather than a final authority.

What changed in August 2026

According to Shah, the key change is that Microsoft made Copilot Credits the primary billing currency across Copilot Studio capabilities. As a result, consumption is now tied more closely to actual agent activity, and billing can start during the build phase rather than only at publish time. This alters planning because teams can no longer assume zero cost while developing complex agents.

Furthermore, Shah notes that licensing now separates access (who can build and manage agents) from usage (how much the agents cost to run). Consequently, organizations must choose both the right access licenses for creators and the correct credit procurement strategy for runtime and authoring costs. This split increases flexibility but also raises the need for tighter cost controls.

How Copilot Credits work

Shah breaks down a Copilot Credit as the unit consumed when agents perform tasks such as summarizing, answering questions, or executing actions. He highlights three main drivers of consumption: the chosen harness, the depth of interaction, and the specific features or tools an agent calls. Therefore, design choices early in development — for instance, opting for more tool calls or deeper context — directly increase credit burn.

In addition, the video explains runtime differences between light, medium, and heavy scenarios and shows how model usage, context size, and tool invocation each contribute to overall consumption. Consequently, teams must weigh usability and intelligence against ongoing credit costs when designing agents. Shah underscores that manual, non-LLM configuration can drastically reduce consumption because it consumes zero credits.

Purchase options and pricing tradeoffs

Shah outlines four ways to obtain credits: a Pay-As-You-Go meter, a Copilot Credit capacity pack, a Copilot Credit Pre-Purchase Plan (P3), and a Microsoft Agent Pre-Purchase Plan. He gives concrete numbers: a capacity pack is described at $200 per month for 25,000 credits and notes that pay-as-you-go charges are around $0.01 per credit. Therefore, organizations must balance immediate flexibility against the lower unit cost of committed plans.

Moreover, Shah explains conversion math for commit-based discounts using Copilot Credit Commit Units (CCCUs), where $1 converts into 100 credits and larger pre-purchases unlock up to a 20% discount. He also highlights the Agent Commit Units (ACUs) used by the Microsoft Agent P3, which can be shared across Copilot Studio, Foundry, Fabric, and GitHub Copilot. Consequently, teams with broad platform use may find pooled commit plans economical, while smaller or unpredictable workloads might prefer pay-as-you-go despite higher per-credit cost.

Harnesses, consumption patterns, and practical choices

The video describes three harnesses: the GitHub Copilot harness, the Standard harness, and the Copilot Chat harness, and Shah maps how each consumes credits. For example, the GitHub Copilot harness starts charging during authoring and scales with light, medium, and heavy authoring scenarios, whereas the Standard and Copilot Chat harnesses follow different rate cards tied to answers and actions. Thus, choosing a harness is both a technical and financial decision.

Shah demonstrates that the single biggest lever for cost savings is reducing reliance on LLM-driven actions and preferring manual or deterministic configurations when suitable. However, he warns that this tradeoff can reduce agent flexibility and user experience, making the decision context-dependent. Consequently, architects should measure expected usage patterns and pilot different harnesses before committing to large purchases.

Challenges, recommendations, and next steps

Finally, Shah highlights practical challenges: forecasting credit consumption is inherently uncertain, licensing terms change rapidly, and the best procurement choice depends on scale and predictability of usage. Therefore, he recommends starting with small pilots, monitoring consumption closely, and adjusting purchase plans as patterns emerge. Additionally, teams should involve procurement early to evaluate commit discounts versus on-demand flexibility.

In summary, the video by Dhruvin Shah [MVP] offers a clear walk-through of the new Copilot Studio licensing landscape, explains tradeoffs between cost and capability, and provides actionable guidance for teams planning agent development. Nevertheless, organizations should verify the current Microsoft documentation and tailor their approach to balance cost, performance, and user experience.

Keywords

Copilot Studio licensing 2026, Copilot credits explained, Microsoft Copilot pricing 2026, Copilot credits management, Copilot licensing guide A to Z, Harness Copilot 2026 features, Copilot Studio enterprise licensing, how to use Copilot credits