Semantic Kernel pricing guide details
Pricing Overview
Semantic Kernel is a free, open-source SDK from Microsoft, released under the MIT license. It has no licence cost and no paid tier of its own: you add it to a project from NuGet or PyPI and use it without seat counts, usage meters or a sign-up.
What you pay for is everything around it. Semantic Kernel orchestrates calls to a model provider, such as Azure OpenAI, OpenAI or Hugging Face, and that provider bills you for every call at its own rates. Your application also runs somewhere, and that infrastructure is billed by whoever hosts it. Microsoft adds no charge for the SDK on top of either.
One more point belongs here, because it affects cost over time: Microsoft now directs new projects to the Agent Framework. Semantic Kernel still works and still costs $0, but a team starting today should weigh that before building on it.
Plan Comparison
There are no Semantic Kernel plans to compare. The bill has three parts, and only one of them comes from Microsoft's SDK:
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Cost component | Price | Who bills you | What decides the amount |
|---|---|---|---|
| Semantic Kernel SDK | $0 | Nobody | Free and open source, no paid tier |
| Model calls | Provider's rates | Your model provider | Which models you call and how many tokens they process |
| Infrastructure | Host's rates | Your cloud or hosting provider | Where the application, memory store and any vector search run |
Because the SDK has no published prices beyond $0, this page does not quote Azure OpenAI, search or database rates. Those change often and depend on your region and contract; take them from each provider's own pricing page.
The value case is simple. The framework is free for one developer or for a team of hundreds, and its connectors let you swap model providers without changing your orchestration code. That makes the provider's rate card, not the framework, the number to negotiate.
Hidden Costs and Considerations
- Agent patterns multiply tokens. Planners, function calling and multi-agent flows send intermediate steps back to the model, so one user request can become several billed calls.
- Azure is the deepest integration. Semantic Kernel supports several providers, but its closest fit is Azure OpenAI. Teams on other clouds may need extra engineering time.
- You build the operations layer. Semantic Kernel has no paid hosting or tracing product of its own, so tracing, evaluation and monitoring come from your own stack or a third-party tool.
- Migration risk. With Microsoft pointing new work to the Agent Framework, plan for the cost of a later move.
Cost Estimates by Team Size
The SDK costs $0 at every team size, so the estimate is the same for a solo developer, a team of 5 and a team of 20: $0 for Semantic Kernel itself.
What grows with the team is usage. To estimate a month, take your provider's per-1M-token rates and work out: input tokens (in millions) x input rate, plus output tokens (in millions) x output rate, for each model you call. Then add hosting for the application and any memory or vector store. A solo prototype may run on a laptop with light model usage; a team of 20 in production pays mainly for model calls and always-on hosting. Neither total includes anything from Microsoft for Semantic Kernel.
How Semantic Kernel Pricing Compares
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Tool | Framework cost | Paid offering |
|---|---|---|
| Semantic Kernel | $0, open source | None |
| LangChain | Open-source framework | Per-seat pricing: $0 per seat (Developer), $39 per seat |
| Dify | Community Edition free to self-host | Dify Cloud Professional $59 per workspace per month or $590 per workspace per year; Team $159 per workspace per month or $1,590 per workspace per year |
| CrewAI | Basic is free with 50 workflow executions per month | Enterprise is custom-priced; no paid self-serve price is published |
Semantic Kernel is the only one of the four with no paid layer at all. LangChain and Dify sell hosted tooling on top of a free core, which you may value for tracing or a visual builder. With Semantic Kernel you get the SDK for free and supply those parts yourself.