Contents
Fast answer
Semantic Kernel is free under MIT, and the reason to consider alternatives is direction rather than price. Microsoft's own AutoGen repository names Microsoft Agent Framework the enterprise-ready successor, and Agent Framework is the stated destination for Semantic Kernel too, while the Semantic Kernel introduction page on Microsoft Learn still carries a July 2023 authoring date and a June 2024 update stamp with no mention of it. Agent Framework is the sanctioned move and is also free. Outside Microsoft, LangChain and LangGraph are the broadest ecosystem, with LangSmith Developer free at 1 seat and 5,000 base traces and Plus at $39 a seat a month. Pydantic AI is the pick for typed structured output in Python. The OpenAI Agents SDK is the minimal option. Google ADK is the one with the broadest language coverage, spanning Python, TypeScript, Go, Java and Kotlin.
Facts last verified 2026-08
Quick facts
- Why people leave
- Microsoft positions Agent Framework as the successor; the Learn intro page is dated 2023
- Every alternative here
- Free; the paid layer is observability
- Microsoft Agent Framework
- $0, the sanctioned destination with a published migration guide
- LangSmith
- Developer $0/seat with 5,000 base traces; Plus $39/seat/mo with 10,000
- Logfire
- Personal free with 10M records; Team $49/mo
- Google ADK languages
- Python, TypeScript, Go, Java and Kotlin
- Hosted agent runtime
- Not verified on a primary pricing page on 2026-08-20
On this page
What this page covers
A comparison visitor should understand the tradeoff, the best-fit scenario, and the next diagnostic tool to confirm the choice.
- 01Why people leave
- 02LangChain
- 03Pydantic AI
- 04LangGraph
- 05OpenAI Agents SDK
- 06Google ADK
Why does this matter now?
Nobody migrates a working system because a framework is free in a different way. They migrate because the documentation is stale, the language they use has stalled, and the vendor has published a successor. Those are all real signals about future maintenance cost, and they are the only currency that matters when the licence fee on every option is zero.
Internal path
Where to go next from this page
These links are part of the A8gent learning and conversion path. Use them to move from concept, to diagnosis, to workflow build, to course.
What you should be able to do after this
- Decide whether to migrate or stay put
- Find a framework with a settled direction
- Match the choice to your language constraints
- Keep Semantic Kernel where it already works
How do you do it, step by step?
1. Microsoft Agent Framework, the sanctioned destination
Free, and named the enterprise-ready successor in Microsoft's own AutoGen repository, with a published migration guide. For a .NET team already embedded in the Microsoft ecosystem, this is the path of least resistance and the one Microsoft's documentation will keep current. Budget the migration as engineering weeks rather than a dependency bump, because the abstractions changed.
2. LangChain and LangGraph, for the broadest ecosystem
LangChain is the widest set of integrations and published answers, and LangGraph is its graph-first runtime for stateful, durable agents. Both are open source and self-host free. LangSmith Developer is $0 a seat a month with a 1 seat maximum and 5,000 base traces, Plus is $39 a seat a month with unlimited seats and 10,000, and compute is metered at $1.50 per LCU and $1.00 per LSU.
3. Pydantic AI, when you want typed output in Python
MIT licensed and free, built around models producing reliable structured output your code can act on. Python only, which rules it out for a .NET team but makes it a strong choice for the Python half of a mixed shop. Logfire sits beside it, free on Personal with 10 million records a month hard capped, $49 a month on Team.
4. OpenAI Agents SDK, when you want a smaller surface
A free open-source SDK with no platform fee, seats or execution meter, deliberately minimal. Semantic Kernel's migration documentation shows a plugin needing an attribute, a wrapper, a kernel and a kernel handoff where the successor takes the function directly. If ceremony was your complaint, this is the lightest option on this list.
5. Google ADK, when language coverage is the constraint
Free under Apache 2.0 and available in Python, TypeScript, Go, Java and Kotlin, which is broader than anything else here. It is the natural place to look if Semantic Kernel's Java support was the thing you were relying on. The trade is churn, since ADK 2.0 superseded its own earlier workflow agent patterns.
The trigger is direction, and it is documented
Semantic Kernel is MIT licensed and free with no fee, seats or execution meter, so nobody leaves over cost. The trigger is that Microsoft has positioned a successor. Its AutoGen repository names Microsoft Agent Framework the enterprise-ready successor to AutoGen and directs new users there, and Agent Framework is the stated destination for Semantic Kernel too. That is a documented direction rather than a rumour, and it changes the calculation for anything greenfield.
The second trigger is that the documentation entry point has not kept up. The Semantic Kernel introduction page on Microsoft Learn carries an authoring date of 11 July 2023 and an update stamp of 24 June 2024, and does not mention Agent Framework at all. That is the first page a new developer lands on, describing a product position the vendor's newer material has moved past by roughly two years. Every hour reconciling docs against reality is paid in salary.
The third trigger is ceremony, and Microsoft's own migration documentation makes it visible. In Semantic Kernel, exposing a function to an agent involves a KernelFunction attribute, a plugin wrapper, a kernel and a handoff to that kernel. In Agent Framework the equivalent takes the function directly. That is the shape of every integration you write, so it compounds across a codebase rather than being a one-off style preference.
The fourth trigger applies to a specific group and is worth checking rather than assuming. Microsoft's overview describes C#, Python and Java support, and the three implementations have not moved at the same pace. A team that chose Semantic Kernel specifically for Java should verify the current state of that release line against .NET and Python before planning around it, because support that has slowed shows up as an unfixed bug at an inconvenient moment.
What each alternative fixes
Microsoft Agent Framework fixes direction and stays in the ecosystem. It is free, named the enterprise-ready successor in Microsoft's own repository, supported by a published migration guide, and it is where the documentation effort is going. For a .NET team this is the path of least resistance and the one whose material will remain current. Budget the migration as engineering weeks rather than a version bump, because the abstractions differ rather than just the imports.
LangChain and LangGraph fix ecosystem breadth. LangChain offers the widest set of integrations and published answers, which matters when you hit something unusual and need an existing solution rather than a design exercise. LangGraph is its graph-first runtime for stateful, durable agents. Both are open source and self-host free; LangSmith is $0 a seat on Developer with a 1 seat maximum and 5,000 base traces, or $39 a seat a month on Plus with 10,000, plus $1.50 per LCU and $1.00 per LSU.
Pydantic AI fixes output reliability for the Python half of a mixed shop. It is MIT licensed and free, built around models producing typed structured output your code can act on, with Logfire alongside for observability: free on Personal with 10 million records a month hard capped, $49 a month on Team and $249 on Growth. It is Python only, which rules it out for a .NET team but makes it a strong choice wherever Python is already the language.
The OpenAI Agents SDK fixes ceremony most directly, since it is the lightest thing on this page. A free open-source library with no platform fee, seats or execution meter, deliberately minimal. And Google ADK fixes language coverage, being free under Apache 2.0 across Python, TypeScript, Go, Java and Kotlin, which makes it the only option here with published Java support if that was your reason for choosing Semantic Kernel in the first place.
Nothing is broken, so time the move rather than rushing it
Be precise about what the successor position does and does not mean. Semantic Kernel is not formally in maintenance mode the way AutoGen is, and it continues to ship releases. An existing codebase that depends on it and works is not broken by Microsoft naming a successor, and migrating for direction alone is engineering time spent for no capability gain. The right plan is usually to schedule the move for the next time you are making substantial changes anyway.
For a new project the calculation is genuinely different and much simpler. Microsoft points new users elsewhere, the successor is also free, and adopting Semantic Kernel today means adopting the thing you will be migrating off. There is no cost saving to weigh against that, because both cost nothing, so the decision comes down to which one your team will still be productive on in eighteen months.
What you keep by staying is worth naming fairly. Semantic Kernel is genuinely well suited to embedding AI into an existing application, with plugins wrapping code you already have, and it is middleware over providers rather than a provider itself, so Azure OpenAI, OpenAI, Anthropic and local models all remain reachable and bill you directly. Nothing flows to Microsoft's framework layer at any volume.
One thing not to budget on either path: the managed hosting destination. We could not confirm published per-hour compute rates for Microsoft's hosted agent runtime on a primary pricing page when we checked on 20 August 2026. If a plan depends on managed hosting rather than your own container, treat that as an unpriced line and get the rate in writing, because running the code in a container you already know how to budget is the route with a forecastable bill.
At a glance
Semantic Kernel alternatives, read from each project's own pricing or documentation page on 20 August 2026
| Alternative | Published price | Reason to switch to this one |
|---|---|---|
| Microsoft Agent Framework | $0 | Named the enterprise-ready successor in Microsoft's own AutoGen repository |
| LangChain and LangGraph | Libraries free; LangSmith $0 or $39/seat/mo; $1.50 per LCU, $1.00 per LSU | Broadest integration ecosystem and a graph-first runtime for stateful agents |
| Pydantic AI | $0, MIT licensed, Python only; Logfire free or $49/mo Team | Reliable typed structured output for the Python half of a mixed shop |
| OpenAI Agents SDK | $0, free open source | Least ceremony; the successor takes a function directly where SK needs wrappers |
| Google ADK | $0, Apache 2.0 | Only option here with published Java support alongside Python, Go and Kotlin |
| Semantic Kernel (for reference) | $0, MIT licensed, C#, Python or Java | Learn introduction page authored 2023-07-11, updated 2024-06-24 |
What mistakes should you avoid?
- Migrating a working Semantic Kernel codebase urgently when nothing is broken
- Starting greenfield work on a framework Microsoft has positioned a successor to
- Trusting the Learn introduction page as current when it carries a 2023 authoring date
- Assuming the C#, Python and Java implementations move at the same pace
- Choosing a Python-only alternative for a .NET team
- Budgeting a hosted agent runtime rate that is not published
FAQ
- What should I use instead of Semantic Kernel?
- For a Microsoft-centred team, Microsoft Agent Framework, which Microsoft's own AutoGen repository names the enterprise-ready successor and supports with a migration guide. Outside Microsoft, LangChain and LangGraph for ecosystem breadth, Pydantic AI for typed output in Python, or the OpenAI Agents SDK for a minimal build.
- Is Semantic Kernel deprecated?
- Not formally deprecated, but Microsoft positions Agent Framework as its successor and the Semantic Kernel introduction page on Microsoft Learn still carries a July 2023 authoring date and a June 2024 update stamp with no mention of it. For a new project that is the material fact; for an existing one it is a reason to plan rather than to panic.
- Do the alternatives cost money?
- The frameworks do not. Microsoft Agent Framework, LangChain, LangGraph, Pydantic AI, the OpenAI Agents SDK and Google ADK are all free. The paid layer is observability, at $39 a seat a month for LangSmith Plus and $49 a month for Logfire Team, both with free tiers underneath.
- What about the .NET and Java options?
- For .NET, Microsoft Agent Framework is the natural successor and stays in the same ecosystem. For Java, Google ADK is the only framework in this comparison with published Java support alongside Python, TypeScript, Go and Kotlin. Most of the popular alternatives are Python-first, which is the real constraint for a non-Python shop.
- When should I keep Semantic Kernel?
- When you have a working codebase that depends on it and no substantial changes planned. It is MIT licensed and free, so there is no cost pressure to move, and migration is engineering time you would spend for direction rather than for capability. Plan the move for the next time you are making major changes anyway.
Sources & further reading
Done for you
Want this built for your business?
Tell us the workflow. We scope, build, and ship the agent with guardrails and the numbers to prove it worked. The scoping call is free.
Free scoping call. You own the code.
Was this page helpful?

