June 2025 Round Up

New SwitchBot Hub With Matter

When I initially tested SwitchBot’s entry into Matter with their Hub 2 device I was a little unimpressed. It lacked polish and didn’t support many of their devices. Since then, they’ve done a great job in upgrading their support so most of their Bluetooth devices now work via the bridge and can be integrated seamlessly into supporting platforms, which is especially good for Apple Home users who were locked out of native support.

The Hub 2 has since become a handy accessory, not just for its hub functions, but as a very capable room climate sensor as well. SwitchBot is now upping the ante on this approach with the new SwitchBot Hub 3.

This new iteration adopts a more modern look, with a sharp IPS display and more physical controls, including a large rotary dial for precise control of supported functions. The dial can be used with SwitchBot devices such as AC controls and lights to provide precision adjustment of temperature and brightness, while the capacitive touch controls can be programmed in the SwitchBot app to trigger scenes with a tap.

Not only can you take control of your SwitchBot Bluetooth devices, but they’ve also included an IR blaster to provide smart control over a huge range of infra-red remote-controlled appliances. Pretty much anything is supported here thanks to the ability to learn the IR signals from the original remote. On the Matter side, the hub also supports up to 30 SwitchBot or third-party Matter devices, so for those without a Matter controller yet you can use this to fit the bill. Importantly, thanks to local control via Wi-Fi, Bluetooth and IR, all of this functionality works without an internet connection.

The SwitchBot Hub 3 is compatible with Amazon Alexa, Google Home, Apple Home, and Samsung SmartThings. SwitchBot is running an introductory promotion with 15% off at the moment. Check it out at the SwitchBot store. If you need to catch up on Matter, check out with my enthusiast guide.

Can AI Models Actually Reason?

Generative AI certainly gets all the attention - and controversy - but AI itself can actually perform a whole bunch of genuinely useful things that don’t involve ripping off creatives’ hard work. Learning languages, data analysis, healthcare, and yes, smart home control. The ability of our voice assistant to actually understand nuance and intent makes this experience far better and more natural than having to recite specific commands syntax.

Being concerned that my voice commands won’t do what I need is certainly a disincentive to even trying, and where that confidence is lacking - especially in front of other people - I’m inclined just to pull out the app and do what I need directly. The idea that our AI assistants can actually understand what we mean makes claims from big AI brands that their models can now ‘reason’ pretty interesting.

Of course, those of us following this tech closely take these claims with a pretty big dose of salt. Large language models simply don’t work in a way that lends itself to any kind of complex thought process, so it seems a stretch to make these claims without a significant shift in the approach. According to new research from Apple’s AI team, our skepticism may be well placed.

The researchers acknowledge that the new models from the likes of OpenAI, Anthropic and Google do show significant improvements on reasoning benchmarks, there are still a lot of unanswered questions. Specifically, the team put these models through their paces in a battery of “controllable puzzle environments”, that is, tests that can’t be solved by looking at existing source material.

Unsurprisingly, these tests showed that the current crop of LRMs (Large Reasoning Models) suffer from complete accuracy collapse beyond a certain puzzle complexity. The models suffer from a “counter-intuitive scaling limit” and begin to ‘overthink’. This is reminiscent of the trend for these new models to have higher rates of hallucination, rather than showing improvements in accuracy, which certainly suggests these companies are going in the wrong direction in their race to victory.

Of course, Apple’s seeming struggle to catch up in the AI race could be blamed for sowing negative publicity against the competition, but these findings are not particularly surprising and may, in fact, be why Apple is treating more carefully.

Wemo’s HomeKit Doorbell Tested

Users of Apple Home often want to take advantage of the privacy-centric HomeKit Secure Video feature when adding security cameras to the home. This feature offers a range of cutting-edge features that generally work very well and completely avoids the risks inherent in having internet-connected cameras and their recordings potentially accessed by nefarious third-parties.

There are only hand handful of video doorbells that support this feature, Wemo’s Smart Video Doorbell being one of them. I’ve put this model through my extensive 32-point testing process to see how it performs under real-world conditions and the results were mixed.

On the one hand, the use of a 3:4 aspect ratio camera ensures HomeKit Secure Video is able to see and identify packages with a high degree of accuracy anywhere in front of your door. On the other hand, poor camera hardware and slow processing result in pretty poor image quality, slow notifications, and missed events.

If package security is your primary focus, then this is your best option for native HomeKit support, otherwise, the Aqara G4 (full review) performs better in every other test category under the same conditions.

See all the test results for yourself, or see my in-depth comparison with the Aqara G4 for the pros and cons.

Eve Energy with Matter

OK, so I’m late to the party here. I’ve actually be using the latest Eve Energy smart plug for a long time now, but I’ve never written a full review on it. I have developed a comprehensive scoring methodology for smart plugs, but I’ve been focused on building up my video doorbell test library, so I haven’t got to many plugs yet. While waiting for my latest doorbell victim to arrive I decided this would be a good time to run the Eve Energy through its paces - officially.

I had really good experiences with Eve products and, unsurprisingly, I found the Eve Energy to be a stellar performer. Thanks to Thread it’s both easy to set up on a variety of smart home ecosystems, super-fast, and extremely reliable.

Eve also doesn’t run their own servers, so their app (which you do need to access energy monitoring and some optional features), does NOT need you to sign up for yet another account. This means it not only performs well but beats most of the competition when it comes to privacy and security.

There’s not much to complain about with this one, but non-US versions do tend to be on the bulky side which means blocking adjacent sockets. That’s frustrating, but not unusual outside of the typically more compact US plugs.

Check out my findings for yourself in this comprehensive test report.