Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11The site operator arranges hosting and delivery of the model file and may be billed for it by its hosting or CDN provider. The visitor pays in a different currency: mobile or metered data, local disk space, waiting time, and the device resources used to run the model. Neither bill is fixed. The operator’s depends on the plan, the provider’s billing rules, traffic, and geography. The visitor’s depends on their connection and device. The “about 50 MB” in the headline is an example size, not a standard for browser AI tools.
What the 50 MB figure does and does not establish
The number comes from a DEV Community article that describes a model of roughly 44 MB plus a 5.95 MB runtime download and setup. That page could not be opened when this piece was checked, so the breakdown should be read as the author’s stated example, not an independently confirmed figure. Other sources show that the size varies widely:
As an Amazon Associate I earn from qualifying purchases.
- Chrome’s built-in model. Google Chrome Help says the exact size of Gemini Nano may change as the browser updates it. The browser’s own download requirements are a different number: Google Chrome Help lists approximately 20 GB of free disk space as an eligibility requirement, and Chrome for Developers’ Prompt API guidance says at least 22 GB of free volume space. Neither figure is the size of the model file.
- A WebGPU avatar bundle. bitHuman’s WebGPU documentation, updated 2026-10-04, describes an avatar web bundle of 50–200 MB. That is one vendor’s avatar feature, not a typical language model.
So the first visit can be tens of megabytes or several hundred, depending on what the tool bundles. The payment question is the same in each case; only the amounts change.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Follow the bytes before following the money
The cost question is easier to answer once the path of a single download is laid out:
#1 Best Overall
- EVOLUTION AMD RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
- The visitor’s browser requests the model file from the site’s origin server or from a CDN in front of it.
- The provider serves the file. The provider’s billing system records the transfer, the requests, and any storage the file occupies.
- The bytes cross the visitor’s internet or mobile connection. The visitor’s ISP or carrier handles that leg and may meter it.
- The browser stores the file locally so later visits can reuse it, if the site’s caching and the browser’s storage rules allow it.
- Inference runs on the visitor’s device, using its CPU, GPU, memory, and battery. No model call is made to the operator’s server for that step.
Each step produces a different cost, and each is paid by a different party. The operator pays for steps 1 and 2 through its provider. The visitor pays for steps 3 to 5 through their own connection and hardware.
Who pays what
| Party | Costs it bears | What sets the size | Source for the cost category |
|---|---|---|---|
| Site operator | Storage, requests, bandwidth or egress, or a flat hosting or CDN plan fee; engineering time to set up delivery | Provider and plan, traffic volume, delivery geography, cache behavior | Google Cloud, AWS, and Cloudflare billing documentation |
| Visitor | Mobile or metered data, download wait time, disk space, device compute and battery | Connection type, file size, how often the file is fetched again | Chrome for Developers; Google Chrome Help |
| CDN or cache provider | Its own transfer and request charges; may reduce repeat fetches from the origin | Cache hit rate, cache-control settings, provider rules | Provider billing documentation |
| Browser vendor (built-in model only) | Model download and browser updates are managed by the browser; no third-party site hosts the weights | Browser model version and update schedule | Chrome for Developers, Prompt API documentation. The vendor’s own cost is not stated in these sources. |
The operator’s bill depends on the hosting arrangement
No single rule applies across providers. The three documented examples below show why the bill must be checked for the specific service.
Google Cloud Cloud CDN
Google Cloud’s documentation says content served through Cloud CDN incurs bandwidth charges and HTTP or HTTPS request charges. A model file is one large object served many times, so the bandwidth line is usually the one that matters. Requests add up when many small chunks or many visitors are involved.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →AWS CloudFront
AWS documents viewer transfer charges for CloudFront under its usage-based arrangements. It also describes flat-rate plans. Under a usage-based plan, each visitor’s download is a metered transfer. Under a flat-rate plan, the operator’s cost is set by the plan, and the first-visit transfer does not produce a separate per-gigabyte line. The plan terms, not the file size alone, determine which applies.
Rank #2
- Built for Local AI Development: AMD Ryzen AI Halo is designed for local AI development and inference, featuring 128GB unified memory and support for up to 200B parameter models to build and run intensive AI workloads locally.
- 128GB Unified Memory: Features 128GB LPDDR5x unified memory at 8000 MT/s with 256 GB/s memory bandwidth, providing a shared memory pool across the CPU, GPU, and NPU to support larger AI models.
- AMD Ryzen AI Max+ 395 Processor: Features 16 cores, 32 threads, and Zen 5 architecture, paired with AMD Radeon 8060S integrated graphics featuring 40 RDNA 3.5 compute units and an AMD XDNA 2 NPU with up to 50 TOPS.
- Linux AI Developer Platform: Purpose-built for Linux-based AI development with full AMD ROCm software support and preloaded tools, models, and workflows optimized for local AI development.
- Compact, Connected Design: Includes a 2TB M.2 SSD, 10GbE LAN, Wi-Fi 7, Bluetooth 5.4, USB-C connectivity, and HDMI 2.1b.
Cloudflare
Cloudflare’s documentation includes bandwidth within its proxied-domain example and describes free egress for its R2 object storage. Those are documented terms for those products at the time of the sources. Plans and allowances change, so the current pricing page for the specific plan and region should be checked before any budget is set.
The visitor’s burden: data, storage, and waiting
Chrome for Developers, in a 2024-05-14 article by Maud Nalpas, Kenji Baheux, and Alexandra Klepper, says: “AI models can be large, which could lead to a large use of mobile data and device storage.” It is a direct warning to developers, and it names the two costs visitors feel most.
Chrome’s own guidance for built-in models recommends an unmetered connection. A visitor on a capped mobile plan can spend a meaningful share of their allowance on one first visit. A visitor on a metered home connection faces the same question at a smaller scale. Storage is the other cost. The model stays on disk until it is removed, and the free-space requirements cited above show how much room a browser-managed model may need.
Inference itself has no per-use fee in the sources reviewed, but it draws on the device. No quantified figure for the electricity or battery cost of local inference was established in these sources, so the visitor’s power cost cannot be stated as a number.
Rank #3
- EVOLUTION RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
The built-in browser path is different
Chrome’s Prompt API is the clearest example of a browser-managed model. According to Chrome for Developers, Chrome downloads Gemini Nano separately the first time an origin uses the API. The documentation states: “The network requirement is only for the initial download of the model.” It also states: “No data is sent to Google or any third party when using the model.”
These statements describe the documented built-in model. They do not apply to a custom web app that bundles its own weights. In the built-in path, the site does not host the model file, so the operator’s hosting bill for that file does not arise. The visitor still bears the download, the storage, and the device cost.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Caching lowers repeat delivery, not the first transfer
Caching changes who pays for the second and later visits. It does not remove the first download. bitHuman’s documentation describes the pattern directly: “One download: the avatar’s web bundle (50–200 MB) downloads to the browser once, then comes from the cache.” Chrome for Developers recommends responsible serving and caching strategies for models.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
A cache helps the operator only if the browser actually keeps the file and the site’s headers allow it to. If the cache is cleared, the visitor is on a new device, or the file changes, the full download happens again. The operator’s saving is therefore probabilistic, and it depends on how many visitors return on the same device.
What the first-visit transfer adds up to
The table below is arithmetic on the 50 MB example. It counts only first-visit transfers of the model file, using decimal units. It does not include runtime files, repeat visits, requests, or any provider price, because no price for a specific plan or region is established here.
| First visits per month | Model transfer at 50 MB each |
|---|---|
| 1,000 | 50 GB |
| 10,000 | 500 GB |
| 100,000 | 5 TB |
To convert these to money, multiply the outbound transfer by the per-gigabyte rate in the plan that applies, or check whether the plan includes that transfer already.
How an operator can estimate its own bill
- Measure the total first-visit payload: the model file, the runtime, and any other assets the page loads before inference.
- Estimate how many visitors arrive without a cached copy each month. New visitors, new devices, and cleared browser storage all count.
- Check the plan. Find out whether bandwidth is included, metered by the gigabyte, or billed as a flat fee.
- Count requests as well as bytes, especially if the file is split into many chunks.
- Confirm the delivery geography, since the same transfer can be priced differently by region.
- Set cache-control headers so that a returning visitor’s browser can reuse the file, then measure the share of first-visit traffic that drops over time.
Options for reducing the operator’s exposure
- Download on demand. Fetch the model only when the visitor opts into the AI feature, not on page load. This avoids a transfer for visitors who never use the tool.
- Use a smaller model where quality allows. A smaller file reduces both the operator’s transfer and the visitor’s data and storage burden. Quality trade-offs need to be tested for the specific task.
- Use the built-in path where it fits. If the browser’s Prompt API meets the need, the site does not host the weights. The trade-off is dependence on the browser’s model and its requirements.
- Tell visitors what the download is. A visible size and an option to proceed on a metered connection shift the decision to the person who pays for the data.
Who ends up paying
The operator pays for delivery through its hosting or CDN plan, and the visitor pays for the transfer on their own connection, the storage on their device, and the inference on their hardware. Caching reduces how often the operator pays for the same file. In the built-in browser path, the operator hosts nothing and the visitor carries the download. The correct answer for any particular tool depends on whether the site hosts its own weights, which provider and plan it uses, and where its visitors are.
Free tools Windows power users keep installed
One-click scans. No signup required.
Sources: Chrome for Developers, Prompt API documentation and the 2024-05-14 article by Maud Nalpas, Kenji Baheux, and Alexandra Klepper; Google Chrome Help (accessed 2026); bitHuman WebGPU documentation (updated 2026-10-04); Google Cloud Cloud CDN, AWS CloudFront, and Cloudflare billing documentation. Provider prices and plan terms change, so confirm them on each provider’s current pricing page.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
