A modest but meaningful refresh is rolling out to Intel-powered Copilot+ laptops. Microsoft has released KB5079255, a component update that silently bumps the on-device Phi Silica language model to version 1.2602.1451.0. It doesn’t come with a flashy changelog, but it reworks the neural engine that powers local AI experiences, from instant text summarization to offline assistance.

A Quiet Refresh, Measured in Tokens

KB5079255 targets only Copilot+ certified systems that pair an Intel processor with a dedicated Neural Processing Unit (NPU) and run Windows 11 version 24H2 or 25H2. No other hardware is eligible. The update is offered automatically through Windows Update once the prerequisite—the latest cumulative OS patch—is applied. After installation, you can verify the new version in Settings > Windows Update > Update history, where “Phi Silica version 1.2602.1451.0 for Intel-powered systems (KB5079255)” appears.

Phi Silica itself is a compact, Transformer-based language model designed from scratch to offload inference to the NPU. Microsoft’s developer documentation describes it as using speculative decoding to accelerate text generation and keep latency low. While the KB notes stay deliberately vague (“This update includes improvements”), the refresh almost certainly fine-tunes the model’s runtime behavior: better quantization, more efficient NPU utilization, or subtle adjustments to prompt handling. These under-the-hood changes are invisible but directly affect how responsive and coherent local AI feels in everyday use.

Who Feels the Difference? Home Users, Developers, and IT

For most owners of Intel Copilot+ laptops—machines from major OEMs bearing the Copilot+ badge—you will likely notice snappier performance when using features that rely on Phi Silica. That includes the Click-to-Do suggestions that appear when you select text, local summarization of documents or web pages, and the AI agent that helps you tweak Windows settings. These tasks run entirely on the NPU without phoning a cloud server, so they work offline and keep your data private. The update should shave milliseconds off response times and reduce occasional stumbles in long-form text generation.

Power users and developers who build apps against the Windows App SDK’s LanguageModel APIs will see a shifting performance envelope. Speculative decoding—a technique that generates multiple tokens ahead of time and discards mismatches—often gets tuned version to version. You might measure improvements in tokens-per-second or a drop in time-to-first-token. Microsoft hasn’t publicized exact metrics, but re-benchmarking critical paths after the update lands on your dev machines is prudent. The component update model also means you could get further Phi Silica refinements without waiting for a semi-annual OS upgrade.

IT administrators face a different challenge. The update will install itself on eligible endpoints unless you’ve explicitly blocked Windows Update or use a controlled deployment ring. Because Phi Silica is tied tightly to the NPU driver stack and the latest cumulative update, skipping the OS baseline will prevent the component from installing—useful if you need to stage rollouts. However, the component cannot be cherry-picked separately from the regular update flow for most tools; it’s a package that appears in Update history and, in some cases, can be uninstalled. The real work for IT teams lies in validating that key workflows (such as employee use of AI features in Word or Edge) don’t break after the update lands. A handful of test devices across Intel, AMD, and Qualcomm SKUs should be standard procedure.

How Microsoft Built a Separate Track for AI

The modular AI component stack is one of Microsoft’s smarter plays in the Copilot+ era. When the first Copilot+ PCs launched with Qualcomm’s Snapdragon X chips, Phi Silica shipped as the on-device brains for Windows’ new AI experiences. It quickly became clear that to keep pace with rapid AI advances, Microsoft needed a way to update the model independently of the entire OS—much like smartphone makers push app updates. Hence, Phi Silica was packaged as a component that could be refreshed via Windows Update on its own cadence. The approach also allowed per-silicon tuning: Intel, AMD, and Qualcomm NPUs each have different architectures, and a one-size-fits-all model would leave performance on the table. KB5079255 is the latest such tuning for Intel chips, following earlier updates that likely tweaked the model for Qualcomm hardware.

While Microsoft’s public support page remains sparse, outside analyses and prior developer communications indicate that Phi Silica falls into the multi-billion parameter range—reportedly around 3 billion parameters—optimized for on-device quantization and NPU offload. Speculative decoding, the acceleration technique Microsoft has described in its developer documentation, is a prime candidate for per-version optimization. Each refresh can adjust how aggressively the model pre-computes sequences, directly impacting the smoothness of local AI features.

Get the Update and Verify It

For most users, the process is straightforward and hands-off. Follow these steps to ensure you’re on the latest Phi Silica version:

  1. Confirm eligibility: Go to Settings > System > About and verify you have an Intel processor and “Copilot+ PC” branding. Also check that you’re running Windows 11 24H2 or 25H2.
  2. Apply pending cumulative updates: Run Windows Update and install the latest patches. The Phi Silica component won’t appear without the matching servicing baseline.
  3. Check for updates again: After a reboot, navigate back to Settings > Windows Update and click “Check for updates.” If the component is ready for your system, it downloads and installs automatically—no separate action needed.
  4. Verify installation: Head to Update history and look for “Phi Silica version 1.2602.1451.0 for Intel-powered systems (KB5079255).” That confirms the refreshed model is in place.

For enterprises, import the KB package into your test ring via WSUS or Microsoft Update Catalog if you manage approval. Run pilot tests on representative Intel hardware, checking features like Recall, Click-to-Do, and any line-of-business apps that call AI APIs. Keep a snapshot-based recovery image handy; rollback pathways for AI components aren’t always straightforward, and testing a rollback scenario in a lab is wise before broad deployment. Also, maintain a compatibility test suite for critical integrations and exercise API paths after the refresh to catch any regressions early.

The Road Ahead for Local Intelligence

Microsoft’s component train isn’t slowing. We can expect similar KBs for AMD-based Copilot+ PCs soon, along with further Intel-targeted updates as NPU drivers mature. The longer-term picture points to deeper integration of Phi Silica into Windows: expect it to power more agents across the OS, from natural-language search in File Explorer to smarter Assistive features. For developers, the Windows AI platform will likely expose finer-grained controls over model selection and fine-tuning, turning Copilot+ devices into more versatile local inference nodes. Keep an eye on Windows Update history—the next Phi Silica bump might arrive with a bit more fanfare and, hopefully, a more detailed changelog.