The main thing to watch is the date. Microsoft published the KB on November 19, 2026, but it says the update won't be offered to Windows 11 24H2 until that version's next servicing period on or after December 8, 2026. The KB gives no similar delay for 25H2 or 26H2.
What KB5136899 actually changes
Microsoft's description is short. The support page says the update "includes improvements" to the Qualcomm QNN Execution Provider AI component on the three supported Windows 11 releases. That is the whole changelog. There's no list of fixes, no new operators, no benchmark numbers and no CVEs.
Here is what the KB does confirm:
| Item | Detail |
|---|---|
| KB number | KB5136899 |
| Component version | 2.2611.1.0 |
| Applies to | Windows 11 24H2, 25H2, 26H2 (all editions) |
| 24H2 availability | Not offered until the servicing period on or after December 8, 2026 |
| Delivery | Downloaded and installed automatically via Windows Update |
| Prerequisite | Latest cumulative update for the device's Windows 11 version |
| Replaces | KB5128781 |
| Update history entry | Windows ML Runtime Qualcomm QNN Execution Provider Update (KB5136899) |
Section summary: This is a routine servicing update to a silicon-specific AI component. Microsoft doesn't detail what changed, so you shouldn't expect any particular speedup.
What the QNN Execution Provider does
ONNX Runtime splits a model's work among execution providers (EPs). An EP is a hardware-specific backend that knows how to run model operations on a particular CPU, GPU or NPU. Microsoft says the QNN EP uses Qualcomm's AI Engine Direct SDK (the QNN SDK) to build a QNN graph from an ONNX model. A supported accelerator backend library then runs that graph. In plain terms, it translates a generic ONNX model into something Qualcomm's hardware can run.
The ONNX Runtime project describes the same component and notes that the OnnxRuntime QNN Execution Provider can be used on Android and Windows devices with Qualcomm Snapdragon SOC's. Its configuration options include separate backends. Selecting 'libQnnHtp.so' or 'QnnHtp.dll' enables the HTP backend, which is the Hexagon NPU path. CPU and GPU backends are also available.
Which hardware is actually in scope?
"All editions" on the KB page refers to Windows editions, not to every PC. Microsoft's Windows ML execution provider documentation lists the QNN EP requirements as Snapdragon(R) X Elite - X1Exxxxx and Snapdragon X Plus (X1Pxxxxx) systems, each with a Qualcomm(R) Hexagon(TM) NPU with driver version 30.0.140.0 or later. The same documentation says these providers are available on Windows 11 PCs running version 24H2 (build 26100) or greater (depending on device and driver compatibility).
So an Intel or AMD desktop won't gain Qualcomm acceleration from this package. If your Snapdragon laptop has an older NPU driver, check that before blaming the EP for anything.
Section summary: The update matters to Snapdragon X-class Windows 11 PCs with a current enough Hexagon NPU driver. On other hardware it does nothing useful.
One of five EP updates released together
Microsoft's "Release history of AI updates" page shows KB5136899 shipped alongside execution provider updates from the other silicon vendors. For Windows 11 24H2, 25H2 and 26H2, the November 19, 2026 row lists:
- AMD MIGraphX: 2.2611.1.0 (KB5136903)
- AMD Vitis: 2.2611.0.0 (KB5136902)
- Intel OpenVINO: 2.2611.1.0 (KB5136901)
- Nvidia TensorRT-RTX: 2.2611.1.0 (KB5136900)
- Qualcomm QNN: 2.2611.1.0 (KB5136899)
That page also lists a separate Windows 11 26H1 package. Qualcomm QNN for 26H1 also moves to Version2.2611.1.0KB5136904. That matters if you manage mixed fleets. 26H1 is not on KB5136899's applicability list, so those machines will report a different KB number for the same QNN version. The "2611" in the version string looks like a year-month stamp (November 2026). Microsoft doesn't explain the scheme, but it fits the earlier QNN releases WindowsForum has covered: KB5096135 brought 2.2605.2.0 to 24H2 and 25H2.
Section summary: Microsoft now ships updates for all its third-party execution providers on the same day. KB5136899 is the Qualcomm package for 24H2, 25H2 and 26H2, and 26H1 gets its own KB.
How to check whether KB5136899 is installed
You don't download or run anything yourself. The update arrives through Windows Update. To check it:
- Install the latest cumulative update for your Windows 11 version first. Microsoft lists it as a prerequisite.
- Open Settings > Windows Update > Update history.
- Look for Windows ML Runtime Qualcomm QNN Execution Provider Update (KB5136899).
- On 24H2, don't expect the update before the servicing period on or after December 8, 2026. If it's missing before then, nothing is wrong.
For developers: don't mix up the version numbers
Microsoft's Windows ML versioning guidance shows how to read an EP's version from code. You inspect PackageId.Version on providers returned by ExecutionProviderCatalog, and PackageId is null when the EP isn't present. On a development machine you can also run:
Get-AppxPackage MicrosoftCorporationII.WinML.*
Microsoft's sample output shows a package named MicrosoftCorporationII.WinML.Qualcomm.QNN.EP.1.8 with version 1.8.27.0. That is an example from the documentation, not the 2.2611.1.0 version in this KB. Read the package version on your own system rather than assuming the two numbering schemes match.
Microsoft's Windows ML documentation also says compatible EP updates reach apps automatically, so apps can pick up performance improvements and new operator support without being updated. It adds that apps currently have no way to install an updated EP programmatically. That describes how EP updates work in general. It doesn't confirm that KB5136899 adds any specific operator.
Troubleshooting: when the update or the EP won't show up
Microsoft's EP troubleshooting guidance depends on the Windows ML version. With Windows ML 2.5.96 or later, EPs are acquired through the Microsoft Store. Versions 2.4.89 and earlier used Windows Update. KB5136899 itself is delivered through Windows Update. For the Windows Update path, Microsoft lists these common blockers:
- Pending restart: Some upstream updates need a reboot before EP downloads continue. Restart, finish any pending updates, then try again.
- Paused updates: Devices with Windows Update paused can't download EPs. Go to Settings > Windows Update and choose Resume Updates.
- Managed devices: Enterprise policies may block component downloads or restrict Windows Update. Ask your IT administrator to check whether policy is blocking EP delivery.
- Still stuck: File a report in Feedback Hub under Developer Platform > Windows Machine Learning. It collects the logs Microsoft needs.
A newer EP also doesn't guarantee a model will run entirely on the NPU. Microsoft's Windows ML CLI documentation says that having an EP installed doesn't mean every operator, data type or shape is supported, and that graph partitioning and CPU fallback depend on the provider, model and session settings. If a model was already falling back to the CPU, test it again after this update.
The bigger picture
Microsoft now services NPU and GPU acceleration much like Defender definitions: small components that update separately from the OS, on a monthly rhythm. The upside is that Snapdragon PC owners get vendor improvements without waiting for app developers or a feature update. The downside is that "improvements" with no changelog makes regression testing harder for admins and developers. If a model's behavior changes in December, did the app change, the driver, or the EP? Record your EP version next to your benchmarks.
Bottom line: KB5136899 is a low-drama, automatic update for Snapdragon X-class PCs. Install your latest cumulative update, expect 24H2 to wait until December 8, check Update history to confirm, and don't count on any specific performance gain, because Microsoft hasn't described one.
References
- KB5136899: Qualcomm QNN Execution Provider update (2.2611.1.0) | Microsoft Support Microsoft Support · 2026-11-19T00:00:00+00:00
- KB5136899: Qualcomm QNN Execution Provider update (2.2611.1.0) | Microsoft Support support.microsoft.com
- QNN Execution Provider - Qualcomm onnxruntime.ai