Apple’s latest Mac mini packs a significant amount of computing power into an exceptionally small chassis, delivering strong results in everyday workloads, 4K video editing and AI-assisted applications. But when it comes to running large language models locally, the limits of its memory and processing power become more apparent.
The new Mac mini M6 retains the compact design introduced with the M4 generation, while bringing a more powerful processor, faster connectivity and hardware specifically designed to accelerate artificial intelligence workloads.
The biggest change for buyers, however, is the price. The Mac mini M6 starts at €1,079 in Italy, €350 more than the previous M4 model. That represents an increase of almost 48%, making the new generation considerably less attractive as a low-cost entry point into Apple’s desktop ecosystem.
The review unit tested was configured with 32GB of unified memory and 2TB of storage and was used for everyday work, photo editing, 4K video production and a series of experiments with AI models running entirely on the machine.
Compact design, familiar layout
The M6 keeps the same compact aluminium design introduced two years ago. Measuring just 12.7cm on each side and around 5cm high, the computer weighs less than 700 grams and includes its power supply.
The small footprint makes it suitable as a permanent desktop workstation or as a secondary computer alongside a laptop. It can also drive up to three external displays.
Connectivity remains broadly unchanged, with two USB-C ports and a headphone jack on the front, while the rear houses three Thunderbolt 4 ports, HDMI and Ethernet. The standard Ethernet connection has been upgraded from 1Gbps to 2.5Gbps, with a 10Gbps option also available.
Apple’s new N1 connectivity chip brings Wi-Fi 7 and Bluetooth 6. There are still no USB-A ports or SD card reader, while the power button remains underneath the machine.
A major performance step over M4
The M6 is built on a 2nm process and features a 12-core CPU and 12-core GPU, compared with 10 cores for each in the M4.
Apple has also added new neural accelerators to the GPU and upgraded the Neural Engine, with two 16-core units providing a total of 32 cores dedicated to certain machine-learning workloads.
Unified memory remains one of the Mac mini’s key advantages for AI applications because CPU, GPU and other components share the same memory pool. The M6 supports up to 32GB, with memory bandwidth reaching 170GB/s in the 24GB and 32GB configurations.
The 2TB SSD in the test machine also delivered strong results, reaching around 6,541MB/s in write speeds and 6,100MB/s in reads.
One important limitation is that the storage chips are soldered to the logic board. Unlike the previous M4 design, the internal storage therefore cannot be upgraded later, making the initial configuration an important purchasing decision.
Strong benchmark and video results
In Geekbench 7, the Mac mini M6 scored 3,951 points in single-core and 21,816 in multi-core testing, representing gains of roughly 24% and 48%, respectively, over current M4 Mac mini averages.
GPU testing produced scores of 84,997 in Metal and 54,219 in OpenCL.
The machine also performed strongly in Cinebench, recording 216 points in single-core, 1,382 in multi-core and 7,030 in the GPU test.
The results demonstrate how far Apple’s consumer chips have advanced, with the M6 approaching the multi-core performance of older high-end Ultra processors despite being installed in Apple’s smallest desktop.
Video editing provided another strong showing. A six-minute 4K project created in Final Cut Pro, including noise reduction, slow-motion footage, multiple video layers, blur and transparency effects, was exported to ProRes 422 in 2 minutes and 35 seconds.
The same project took around 3 minutes and 45 seconds to export in HEVC. In both cases, the export finished faster than the actual video duration, while the machine remained responsive during the process.
For 4K video production, social media content and moderately complex editorial projects, the M6 offers more than enough performance. Users working regularly with highly complex projects, extensive high-resolution timelines or demanding 3D workloads may benefit from a higher-end configuration.
Local AI reveals the limitations
The biggest question is whether the Mac mini M6 can also serve as a serious local AI workstation.
The answer depends heavily on what “local AI” means.
Running relatively small language models or AI features integrated into everyday software is a very different task from running models with tens of billions of parameters or generating complex images locally.
Geekbench AI showed substantial gains over the M4, particularly on the GPU. The M6 scored 17,833 in single precision, 33,469 in half precision and 35,666 with quantized models, compared with 8,333, 9,601 and 8,798 respectively for the M4.
The real-world results were more nuanced.
Using LM Studio with a 27-billion-parameter Qwen model quantized to 4-bit, the Mac mini was able to run the model and analyse a roughly 2,000-word article before producing a summary, key points and quotations.
However, the complete task took 18 minutes and 10 seconds, partly because the model generated a long reasoning process. A smaller seven-billion-parameter Mistral model reduced the processing time to around eight minutes.
The machine remained usable for other tasks during the tests, but the results show that having enough memory to load a large model does not necessarily translate into professional-level response times.
Image generation is possible, but slow
The same limitations appeared in image generation.
Using Draw Things and a 6-bit quantized version of Qwen Image 2512, the Mac mini generated an image of a technology newsroom in around nine minutes.
The result was broadly convincing, although familiar generative AI errors appeared in finer details.
Nine minutes for a single image may be acceptable for experimentation, but it becomes impractical when producing multiple versions or working against tight deadlines.
AI is already embedded in everyday software
There is another side to the AI equation that may be more relevant to most Mac mini buyers.
AI-powered features are increasingly built into applications people already use, from automatic subject recognition and image noise reduction to detail reconstruction, voice separation and frame generation in video editing.
These workloads do not require users to install and configure large language models themselves. Instead, the applications use specialised models behind the scenes.
The M6’s upgraded Neural Engine and GPU accelerators are designed to handle precisely these types of workloads. How much of the additional hardware performance users actually experience will depend on software optimisation and how effectively developers use the new architecture.
Price makes the decision harder
The biggest drawback of the M6 may ultimately be its price.
The entry-level model in Italy costs €1,079, compared with €729 for the M4 at launch. The base configuration includes 16GB of unified memory and just 256GB of storage.
Moving to 24GB of memory costs an additional €220, while 32GB adds €440. That puts the 32GB version at €1,519 before any storage upgrade.
The fully equipped test configuration with 32GB of memory and 2TB of storage reaches €2,619, bringing it much closer to higher-performance Mac mini configurations based on the M5 Pro.
The Mac mini M6 is therefore an extremely capable compact computer, particularly for everyday productivity, creative work and the growing number of AI features built into mainstream software. It can also run sizeable AI models locally.
But for users whose primary objective is to run large language models continuously and quickly, 32GB of unified memory and the M6 architecture may not be enough to deliver the experience they are looking for. In that scenario, a configuration with more memory and a higher-end processor is likely to be more appropriate.
Be First to Comment