Google unleashes gemini's brainpower – but there's a token catch
Google just flipped a switch, making its advanced Gemini “thinking levels” freely available across web, Android, and iOS. This comes hot on the heels of an $80 billion AI infrastructure investment announcement – a remarkably swift turnaround. But before you dive headfirst into deeper analysis, there's a crucial detail you need to know.
Standard vs. extended: what's the difference?
Previously rolled out during Google I/O 2026, Gemini’s thinking modes offer two tiers: Standard and Extended. The Standard setting delivers straightforward responses, while Extended unlocks more complex reasoning and deeper dives into the model's capabilities. Josh Woodward, Google’s VP of Gemini, confirmed the widespread rollout on X, effectively putting the power in everyone’s hands.
You can easily toggle between these modes within the Gemini settings. The Extended option, however, isn't entirely free. It burns through tokens at an accelerated rate, a point many users discovered—sometimes painfully so—as they watched their Gemini quotas disappear unexpectedly.
The real kicker? Google has tweaked its models to consume varying amounts of tokens based on task complexity. This adjustment, while intended to optimize resource usage, led to a wave of user frustration and reports of rapid token depletion.
Ai ultra: the deep think advantage (for a price)
For those willing to invest, Gemini 3.1 Pro now boasts a “Deep Think” option – a premium tier above Standard and Extended. Exclusive to AI Ultra subscribers (a monthly investment ranging from $100 to $200), Deep Think promises the most in-depth analysis and, theoretically, the best possible query results. But let's be honest, with Large Language Models, the quality of your output is inextricably linked to the quality of your prompt. Get that wrong, and even the most powerful AI can falter.
This expansion comes with a considerable environmental impact. Google’s data centers vent a staggering 3.8 million cubic meters of carbon dioxide annually – dwarfing the carbon footprint of an entire small town, which typically sits around 50,000 cubic meters. The question isn't just about cost, but also about the escalating resource demands of increasingly sophisticated AI models.
A token of caution
While Gemini’s accessibility is undeniably exciting, responsible usage is paramount. Consider subscribing to a paid model if you find yourself frequently leveraging the Extended or Deep Think modes. Google's rapid advancement in AI is impressive, but it’s a race against both resource constraints and user expectations.