Qwen3.6 is an open-weights AI model with 35 billion parameters, designed for various computational tasks including data analysis.

How much space did Qwen3.6 identify for deletion?

Qwen3.6 identified approximately 17GB of data that could be potentially deleted from a full SSD.

Why were some important files missed in the report?

Some deletable items, such as those in the Recycle Bin, were not included due to the limitations of the model's context window and memory constraints.

Is it safe to use AI for disk cleanup?

Using AI for disk cleanup can be beneficial, but it is recommended to have human oversight to ensure no important files are mistakenly flagged for deletion.

What kind of files can be safely deleted according to Qwen3.6?

Qwen3.6 recommended deleting cache files and temporary files, which are generally safe when not needed for current tasks.

Using Qwen3.6 to Clear Space on a Full SSD

An exploration of Qwen3.6's capabilities in optimizing disk space on a full SSD, revealing both potentials and limitations for users.

Dealing with a nearly full SSD can be a hassle as users constantly seek ways to free up space. One inspired tech enthusiast sought to leverage an AI model to assist in pruning his storage, using the model Qwen3.6 to identify files suitable for deletion on a Windows 11 system.

The Challenge of Disk Space

Every few weeks, a 1TB SSD fills up, compelling owners to sift through files and determine what can be deleted. The manual search method, typically involving software like WizTree, can be time-consuming. While WizTree excels in pinpointing large files and folders, it leaves the decision on what to delete up to the user. This prompted the user to wonder if an open-weights AI model could assist, especially in providing recommendations for files that could be safely eliminated.

Enter Qwen3.6

The selected AI model for this task was Qwen3.6-35B-A3B, a sophisticated mixture-of-experts model with 35 billion total parameters and 3 billion active parameters. Running on an RTX 4070 Ti Super GPU with 16GB GDDR6X memory, the model utilized a token context of 32,768. Configuration settings included a sampling temperature of 1.0 and a top-k value of 20.

Methodology and Limitations

This experiment gave Qwen3.6 read-only access to a SQLite database derived from a full scan by WizTree. The model essentially used four parameters: 1) largest files, 2) folder sizes, 3) files within specific folders, and 4) totals by file type. Sensitive information was excluded, ensuring privacy throughout the process. With the model's actions limited to read-only and devoid of any capability to directly interact with files, the user aimed to glean insights into what could be removed without causing harm.

Results: Potentially Deletable Items

After running the model, Qwen3.6 identified approximately 17GB of data that could be deleted. The bulk of this—around 16GB—came from a single rolling cache used by Microsoft Flight Simulator. While this represented a significant potential gain in space, the file would immediately regenerate upon relaunching the game.

Additional Findings

Microsoft Edge caches: Found approximately 496MB, a reasonable candidate for deletion.
Temp folder files: About 150MB was identified as safe for removal.
Duplicate folders: The model erroneously suggested deleting the C:\Program Files\llamacpp folder, which was in use at the time.

Missed Opportunities

Interestingly, some notable files that could have aided in freeing up space were overlooked. For example, the model listed files in the Recycle Bin, which contained around 21GB of data that had already been marked for deletion. Not recommending to clear the Recycle Bin represented a significant oversight. Similarly, it identified a 7.4GB crash report located in the Windows\LiveKernelReports folder and a substantial 6.5GB shader cache in the Nvidia\DXCache folder, both of which were also left out of the final report. These represent a missed opportunity for actual storage reclamation.

A cluttered desktop with multiple folders and the Recycle Bin visible.

Constraints of the Current Setup

The investigations revealed limitations intrinsic to the model's operational context. With the Qwen3.6 model itself being 16.5GB and the available memory of the RTX 4070 Ti Super capped at 16GB, competition for resources was evident. The constraints of maintaining a large context (in this case, 32K) while also ensuring output on performance metrics left some recommendations out of view, leading to the oversight of larger files for deletion.

Suggested Clean-Up Items

Item	Size	Status	Comment
MSFS rolling cache	16GB	Yes	Rolls back when game is launched.
Edge browser caches	496MB	Yes	Safe to delete.
Temp installer leftovers	150MB	Yes	Safe to delete.
Llama.cpp folder	1.1GB	No	Incorrect recommendation; in use.
Recycle Bin	21GB	No	Should have been included.
Crash dump	7.4GB	No	Should have been included.
Nvidia DXCache	6.4GB	No	Should have been included.

Concluding Thoughts

While Qwen3.6 demonstrated an impressive ability to analyze data and report potential deletions without risking harm to the system, it stopped short of offering a fully comprehensive cleanup strategy. Users may find that the application of AI for disk management is feasible, but requiring more rigorous checks and balances will help ensure complete efficiency. Future iterations could utilize a more potent context to prevent missing crucial files and improve user experience.

Key Takeaways

The AI model Qwen3.6 was able to identify around 17GB of deletable data.
Many recommendations were hindered by memory limitations, leading to missed deletion opportunities.
Notably, items in the Recycle Bin were not included in the final report.
For now, human oversight remains crucial in managing disk space cleanup.