I tried to love Gemini Live ahead of its Google Home debut, but I can’t imagine paying for it

With Gemini slowly but finally rolling out to Google’s smart speakers and displays, I’ve found myself filled with renewed optimism for my stack of devices. The search giant has stripped the platform of countless features over the past couple of years, and even some basic voice commands have failed to function like they once did. I’m not alone — hundreds of users have voiced the same complaint about Google Home across social media.

Luckily, Gemini looks like it’s on track to revive Google Home’s usefulness but I’m equally concerned that it will introduce problems of its own. Given that I already have Google’s AI Pro subscription, I’m also looking forward to using Gemini Live — the two-way voice conversation feature that won’t require a wakeword every single time. However, after extensively using it on my phone, my excitement levels for the feature’s expansion have been significantly dampened.

Google markets Gemini Live as a way to get “real-time voice assistance” but what it doesn’t tell you is that you’re getting a vastly different experience compared to the AI in its default state. You see, Gemini Live uses a different model that’s specifically optimized for speed and latency rather than depth or nuance. For the most part, it can answer general questions and even perform basic calculations. But after conversing with Gemini Live for just a while, I’ve found that I can very easily run up against its limits.

Take this example: I asked, “What’s expected to arrive with the next-gen iPad mini? Should I buy the current gen or wait for the next one?” and here’s what Gemini Live said:

Ah, the iPad Mini! Well, the latest generation, which was released in October 2024, has quite a few improvements. It’s got the powerful A17 Pro chip, which means it’s super quick, and it also supports Apple Intelligence. The display is lovely and bright, and it’s compatible with the Apple Pencil Pro. Are you looking to upgrade your current Mini, or is this your first iPad Mini?

Now, this answer might be acceptable if I was just looking to have a friendly conversation about the current-gen iPad Mini. But that wasn’t the case — I specifically asked about a future, upcoming model that has seen various leaks in the news lately. The AI completely sidestepped this question, and responded with more filler than substance. In fact, the response felt more like a marketing spiel than helpful assistance.

I could chalk up Gemini Live’s disappointing response to a technology limitation — there’s an accuracy price to pay for real-time responses, right? Unfortunately not, because I’ve found that ChatGPT’s Voice mode blows it out of the water every single time. Here’s a side-by-side comparison of how the two chatbots responded:

To rule out the possibility of Gemini Live fumbling only as a one-off, I decided to try again with a different, unrelated product. This time, I asked about NVIDIA’s rumored RTX 50 Super graphics cards. And yet again, the AI fumbled and spoke about the already-released, non-Super lineup. When I prodded it further, the model doubled down on its hallucinations and began talking about an unrelated product “expected mid-to-late July 2025.” This means the AI clearly looked up some information online but lacked sufficient grounding to know that the product had already been released and that the date was in the past.

Once again, ChatGPT’s Voice mode was significantly more helpful. The overall answer was shorter, so I didn’t have to listen to the AI voice drone on, and I got significantly more information than Gemini Live. When I went into the ChatGPT app to review the conversation, I was also pleasantly surprised to see the sources used for each response.

In case you haven’t used ChatGPT Voice mode, the way it works is slightly different from Gemini Live. Rather than immediately responding to every single question, it will sometimes take a second or two to look up information first. This is audibly signalled by a short chime that lets you know the AI model is accessing the internet. I much prefer this added latency as it leads to a more informed response compared to Gemini Live’s near-instant but often flawed response. ChatGPT can also deliver instant responses when it doesn’t need to look up additional information, so it’s not worse than Google’s offering in that area.

Google’s general-purpose Gemini text-only models have improved leaps and bounds since their rocky debut. Simply ask Gemini 2.5 Flash the same question as I did above and you’ll find that the model is perfectly capable of answering with accurate details (pictured above). You don’t even need to engage the more capable 2.5 Pro model to get this kind of high-quality response.

Gemini Live feels a bit like a letdown by comparison. Given that the feature has existed for over a year now, I’m surprised that Google hasn’t addressed its most glaring limitation yet. What’s worse is that Gemini Live was unable to look up information via Google Search at launch but the search giant quietly added this capability a few months ago. Despite that, however, the AI model clearly isn’t searching even when it should.

Meanwhile, Google has expanded Gemini Live in other ways with new voices, features like the ability to share your camera feed, and plans to bring it to more devices like the Pixel Buds and smart speakers. I can’t help but feel like it’s putting the horse ahead of the cart here. That’s especially true since the search giant wants us to pay $10 per month to access the feature on our smart speakers.

ChatGPT’s Voice mode delivers a significantly better experience for free, and OpenAI even dropped the subscription requirement to use it once demand levelled out. Of course, you can’t use it on a smart speaker yet so the comparison isn’t entirely fair. But that will change sooner or later.

With Google Home’s precarious reputation, Gemini offers the company a rare redemption arc. But Gemini Live feels like it could become another stumble that frustrates some of the company’s most loyal users. I hope this won’t be the case, but time will tell.