Thank you for reporting an issue with the Prompt API in Microsoft Edge!
What happened?
The LanguageModel API is exposed, but the underlying model execution feature remains disabled.
This was tested in Microsoft Edge Dev 154.0.4251.0, Official build, 64-bit.
This reproduces on both:
The following call returns "unavailable":
await LanguageModel.availability()
The following call rejects:
await LanguageModel.create()
Actual error:
NotSupportedError: The feature flag gating model execution was disabled.
The same warning appears in the DevTools console.
I expected availability to return "downloadable", "downloading", or "available", followed by the model download or successful session creation.
This regression reproduces across multiple machines, including a Surface Laptop 7 and a Microsoft Dev Box. It occurs regardless of whether GPU acceleration is available. The Prompt API worked previously, but the current Edge Dev build now blocks model execution with the feature-gating error.
Device performance class
The page does not display a named High, Medium, or Low value. Edge on-device internals reports:
- GPU blocked
- Foundational Model: Not Eligible
- device capable: true
- enabled by feature: false
- Detected VRAM: Not Available
- Minimum VRAM required: 5500 MiB
The diagnostic values above are from the Microsoft Dev Box. GPU blocking on that machine is expected and is not the cause of this issue. The same feature-gating failure reproduces on other capable hardware, including a Surface Laptop 7. The unexpected state is "enabled by feature: false" even though the required flags are enabled.
Is device capable?
true
Edge channel and version
Microsoft Edge Dev 154.0.4251.0, Official build, 64-bit
Chromium version: 154.0.8025.0
Revision: 7063dc1525c44930ae89d34ac9f4a5ba7a07fd2d
Device information
Windows 11 Version 25H2, build 26200.9106, x64
GPU information
The following GPU details are from the Microsoft Dev Box, where GPU access is intentionally disabled. They are included for completeness and are not required to reproduce the issue:
ANGLE (Microsoft, Microsoft Basic Render Driver (0x0000008C) Direct3D11 vs_5_0 ps_5_0, D3D11-10.0.26100.8875)
DirectML feature level: 5.0
Enabled flags
- Prompt API for on-device language model
- edge-llm-prompt-api-for-phi-mini
- Enable prerelease on-device language model
- edge-llm-on-device-model-prerelease
- Prompt API Tool Use
- edge-llm-prompt-api-tool-use
- Enable prerelease on-device AI model tool-use workaround
- edge-llm-on-device-model-prerelease-tool-use-workaround
Edge was fully restarted after enabling these flags.
The effective command line includes:
AIPromptAPI
AIPromptAPIStructuredOutput
AIPromptAPIToolUse
msEdgeLLMOnDeviceModelPrerelease
msEdgeLLMOnDeviceModelMuffinToolUseWorkaround
Enterprise policy
The BuiltInAIAPIsEnabled policy is not configured. The ExperimentationAndConfigurationServiceControl policy is FullMode, value 2.
Additional control test
I reproduced the same failure in a separate Edge Dev profile after explicitly enabling Chromium's OptimizationGuideOnDeviceModel feature on the command line. The internal status still reports "enabled by feature: false".
This report is about a general model-execution feature-gating regression. It reproduces on multiple machines regardless of GPU availability.
Thank you for reporting an issue with the Prompt API in Microsoft Edge!
What happened?
The LanguageModel API is exposed, but the underlying model execution feature remains disabled.
This was tested in Microsoft Edge Dev 154.0.4251.0, Official build, 64-bit.
This reproduces on both:
The following call returns "unavailable":
The following call rejects:
Actual error:
The same warning appears in the DevTools console.
I expected availability to return "downloadable", "downloading", or "available", followed by the model download or successful session creation.
This regression reproduces across multiple machines, including a Surface Laptop 7 and a Microsoft Dev Box. It occurs regardless of whether GPU acceleration is available. The Prompt API worked previously, but the current Edge Dev build now blocks model execution with the feature-gating error.
Device performance class
The page does not display a named High, Medium, or Low value. Edge on-device internals reports:
The diagnostic values above are from the Microsoft Dev Box. GPU blocking on that machine is expected and is not the cause of this issue. The same feature-gating failure reproduces on other capable hardware, including a Surface Laptop 7. The unexpected state is "enabled by feature: false" even though the required flags are enabled.
Is device capable?
true
Edge channel and version
Microsoft Edge Dev 154.0.4251.0, Official build, 64-bit
Chromium version: 154.0.8025.0
Revision: 7063dc1525c44930ae89d34ac9f4a5ba7a07fd2d
Device information
Windows 11 Version 25H2, build 26200.9106, x64
GPU information
The following GPU details are from the Microsoft Dev Box, where GPU access is intentionally disabled. They are included for completeness and are not required to reproduce the issue:
DirectML feature level: 5.0
Enabled flags
Edge was fully restarted after enabling these flags.
The effective command line includes:
Enterprise policy
The BuiltInAIAPIsEnabled policy is not configured. The ExperimentationAndConfigurationServiceControl policy is FullMode, value 2.
Additional control test
I reproduced the same failure in a separate Edge Dev profile after explicitly enabling Chromium's OptimizationGuideOnDeviceModel feature on the command line. The internal status still reports "enabled by feature: false".
This report is about a general model-execution feature-gating regression. It reproduces on multiple machines regardless of GPU availability.