Surface Pro and Surface Laptop Lead DXOMARK Webcam Rankings With On-Device AI

Here's what Microsoft claims, what DXOMARK's own tests add, and what IT buyers should take from both.

The short version​

• New AI camera features: Microsoft says AI Noise Reduction Gen2 and AI Enhancement are available on select Surface Pro and Surface Laptop configurations with Snapdragon X2 Series processors.
• Ranking claim: Microsoft paid DXOMARK to test the devices in April 2026. A Surface Laptop 13.8-inch and a Surface Pro 13-inch, both with Snapdragon X2 Plus (10 core), 16 GB of RAM and 256 GB of storage, took first and second place in DXOMARK's laptop camera ranking as of June 16, 2026.
• Processing stays on the device: Microsoft says the camera AI models run entirely locally on the NPU. Image and video data is not sent to the cloud for this processing.
• Limits on the claims: The ranking applies only to devices with the same camera hardware, firmware and settings as those tested. Microsoft says rankings may change and that results vary by configuration, settings and video-conferencing app.

What DXOMARK found​

This is where the story moves beyond Microsoft's own account. DXOMARK's published review of the Surface Laptop 13.8-inch says it achieves the highest score in our ranking with an outstanding result of 146, surpassing the Apple MacBook Pro 14'' (M4, 2024). The lab credits it with accurate exposure, natural color reproduction, and reliable focus behavior across a wide range of usage scenarios, and says it holds accurate and stable target exposure in all tested lighting conditions, especially in backlit scenes.

The Surface Pro 13-inch scored 144 and came second. DXOMARK says it secures the second position in our ranking, ahead of the Apple MacBook Pro 14'' (M4, 2024). Its subscores are also worth a look. The Surface Pro has the best exposure subscore on the list at 93. The Surface Laptop 13.8-inch has the best color subscore at 84. The Surface Pro scored 40 for texture against a best of 56 (Lenovo ThinkPad X9 Aura) and 76 for noise against a best of 85 (Apple MacBook Pro 14-inch with M5). The same leaderboard appears on other DXOMARK review pages.

DXOMARK's review of the Surface Pro also lists weaknesses:

• Detail is slightly limited, and texture sometimes looks unnatural, especially in low light
• Noise is visible in low light and when there is motion
• Some artifacts appear, including anamorphosis and quantization

So these are the two best laptop webcams DXOMARK has tested, but they are not flawless. Low light remains the hardest case, and that is exactly the problem AI Noise Reduction Gen2 is meant to address. DXOMARK says its method combines objective measurements with more than 20 hours of perceptual evaluations under controlled lab conditions and real-life scenarios. That is a decent protocol. Still, Microsoft commissioned the tests, and readers should keep that in mind.

How this compares with the previous generation​

This isn't Surface's first strong result with DXOMARK. In August 2025 the lab reviewed the earlier Surface Laptop 13-inch and wrote that for the first time, a Windows effectively challenges the long lead of Apple in terms of camera quality. That review also noted that noise is visible in most tested conditions. A year later, DXOMARK describes noise on the new Surface Pro as well controlled in most situations and a problem mainly in low light and motion. That fits Microsoft's account of steady improvement across generations. DXOMARK has since changed how it presents laptop results, though, so older and newer numbers shouldn't be compared directly.

Section summary: An outside lab backs up the ranking. It does not verify any of Microsoft's internal engineering claims, which are covered next.

Inside the pipeline, as Microsoft describes it​

Microsoft splits the camera system into three layers.

1. The camera module. Microsoft says the module is purpose-built for Surface. It combines lens, sensor, infrared filter and electronics, and assembling it takes micrometer-level precision. It also has to fit the device's design and use as little power as possible.
2. The image signal processor (ISP). The ISP turns raw sensor data into the picture you see on a call. Microsoft says it has tens of processing blocks and thousands of adjustable parameters covering contrast, tone mapping, aberration correction, color, sharpness and noise.
3. In-house AI models. These run on the NPU and handle problems that conventional processing struggles with.

The ISP parameters affect each other. Microsoft gives the familiar examples: more sharpening makes noise easier to see, heavy noise reduction wipes out real detail, and brightening a face changes how the background looks. The company calls tuning a "multivariable optimization problem." It says it balances these trade-offs using modeling, lab testing, field testing and psychovisual analysis, which means studying how people actually perceive the image.

A hybrid pipeline, not AI everywhere​

Microsoft doesn't route every frame through a neural network. It says the ISP handles hundreds of standard operations, and AI is used only where conventional processing falls short. Its example is telling a small, distant fly apart from a dead pixel, sensor noise or another artifact. Each task goes to the ISP or to AI depending on its image-quality and power needs.

That matters for battery life. Heavier AI processing uses more energy, and a laptop webcam runs for hours of meetings every day.

The three AI features​

Microsoft stresses that the models work from the original image and video signal and don't generate artificial textures. In other words, it says they clean up the image rather than invent detail. That fits with DXOMARK's comment that texture on the Surface Pro can occasionally look unnatural. Keeping processing faithful and making it look natural are not the same thing, and low light is where the two are hardest to reconcile.

The NPU numbers​

Microsoft gives more specific figures than a hardware blog usually does:

• AI Noise Reduction Gen2: 120 billion operations per frame, 586,000 parameters
• AI Enhancement: 70 billion operations per frame, 132,000 parameters
• Memory footprint: less than 8 MB for "the models." Microsoft doesn't say whether that means stored weights, runtime memory or something else.
• Throughput: Microsoft says the team demonstrated full-frame input and output at 30 frames per second on the NPU with no dropped frames.

The post doesn't describe how that demonstration was run, so treat these as company-reported figures, not independent benchmarks. Operations per frame also tell you nothing about image quality. A bigger number is not a better picture.

The architecture point is more useful. Microsoft says NPU designs differ between chip platforms, so each model is tuned for its target NPU. The processing runs continuously in the background, and nothing needs to be switched on. That is a better use of the Copilot+ era's NPUs than yet another chatbot button.

The earlier note in the post about 'millions of parameters' seems to contradict the per-model figures of 586,000 and 132,000. Microsoft doesn't explain the difference. It may refer to all the models across different camera systems, but that is our guess.

What IT admins and buyers should know​

1. Check the exact configuration. The ranking comes from specific X2 Plus, 16 GB/256 GB configurations. Microsoft says the AI features are on "select" Snapdragon X2 configurations and doesn't publish a full list. Confirm support for the exact SKU with your reseller or Microsoft rep before a purchase depends on it.

2. Watch the model names. The blog's opening mentions the Surface Pro 12-inch and Surface Laptop 13-inch, but the DXOMARK footnote covers the Surface Pro 13-inch and Surface Laptop 13.8-inch. Microsoft's own Surface Laptop product page words its footnote slightly differently. It cites commissioned testing by DXOMARK in April 2026 of Surface Laptop, 13.8-inch with Snapdragon® X2 M10 16GB/256GB. It's probably the same device under a different processor name, but purchase orders should use exact names.

3. Keep firmware consistent. The ranking holds only for the tested firmware and settings. Camera behavior can change with driver and firmware updates, so standardize updates across the fleet through your usual Surface driver and firmware channels, such as Windows Update, Intune or MSI packages. Otherwise you may get uneven results between identical machines.

4. Your meeting app matters. Microsoft's disclaimer names the video-conferencing platform as a variable. Teams, Zoom and others apply their own processing, such as background blur and compression, on top of the camera output. Test in the apps your staff actually use.

5. On-device processing is good for privacy. Local processing means camera frames aren't sent to the cloud for enhancement, which is simpler for regulated industries. That is Microsoft's statement, not an independent audit.

Our take​

Microsoft's post is a good explainer on why webcam quality is a systems problem: sensor, ISP, firmware, AI, power limits and industrial design all have to work together. Apple has been building MacBook cameras this way for years, and DXOMARK's leaderboard shows the Surface team has caught up and pulled ahead in the lab's video-call scoring.

Still, the company that wrote the explainer also paid for the benchmark, and DXOMARK itself says low light and motion remain weak spots. If your staff take calls in dim hotel rooms, the AI noise reduction will be working hard, and you should test it yourself before you trust a slide deck.

Have you used an X2-based Surface Pro or Surface Laptop on Teams in bad lighting? Tell us in the forums how AI Noise Reduction Gen2 holds up.