With Google I/O 2026 almost here, Google seems unable to stop Gemini leaks from slipping out early. Every other day, something new appears inside the app, and this time it looks like Google is experimenting with giving users more control over how much “thinking” Gemini actually does before responding. The new feature, dubbed “Thinking Level,” could fundamentally change how users interact with the AI, allowing them to adjust the depth of reasoning for each query.
According to early reports, some users are now spotting a new “Thinking Level” option inside the Gemini app. The feature reportedly appears within Gemini’s existing model picker, where users already choose between options like Fast, Thinking, Pro, or Google AI Plus. Instead of simply choosing which model you want, Google also appears to be testing how deeply that model reasons through a task. The report says the new Thinking Level option currently shows up when selecting Fast (Gemini 3 Flash) or Gemini 3.1 Pro with thinking enabled. For now, the rollout seems extremely limited, likely confined to a small percentage of users for early feedback.
Things are getting a little more nuanced
If this sounds familiar, that is because Google AI Studio already offers similar controls with Low, Medium, and High reasoning levels. Bringing that flexibility into the regular Gemini app feels like the next obvious step, especially as AI companies increasingly compete on how “thoughtful” or agentic their assistants can feel. The move reflects a broader industry trend: giving users granular control over AI behavior. OpenAI, for instance, recently introduced a “reasoning effort” parameter in its API, and Anthropic has long offered temperature and top-p controls. However, Google’s approach is more user-friendly, embedded directly in the app interface rather than requiring developer-level tweaks.
The Thinking Level feature essentially allows Gemini to allocate more or fewer computational resources to analyzing a query. At the lowest level, the AI might generate a quick, surface-level response based on pattern matching. At the highest level, it could engage in multi-step reasoning, checking its own outputs and exploring alternative solutions before settling on an answer. This is particularly useful for complex tasks like mathematical proofs, logical puzzles, or detailed research questions where accuracy matters more than speed. Conversely, for everyday queries like “What’s the weather today?” or “Set a timer for 10 minutes,” a lower thinking level can save time and battery life on mobile devices.
Honestly, I feel this could be more useful than it sounds
Not every AI request needs maximum reasoning power. Sometimes you just want a quick answer without waiting several extra seconds while the model overanalyzes your grocery list, as if it were preparing a PhD thesis. Giving users control over that balance between speed and deeper reasoning could make Gemini feel much more flexible day to day. For power users, the high thinking level might enable Gemini to tackle problems that previously required specialized AI tools. For casual users, the low level ensures the assistant remains snappy and responsive. This tiered approach also aligns with Google’s broader vision of making AI ubiquitous: by letting the user decide how much “brainpower” to apply, the assistant can adapt to context without manual model switching.
From a technical perspective, the Thinking Level likely adjusts the number of tokens used in the chain-of-thought reasoning process. Gemini’s architecture supports chain-of-thought prompting, where the model explicitly lists intermediate steps before arriving at a conclusion. More thinking means longer chains and more self-correction loops. Google has not disclosed exact parameters, but similar systems in AI Studio offer three presets: Low (fast, minimal reasoning), Medium (balanced), and High (thorough, multi-check). The app version may adopt the same scale or offer a slider. Early UI screenshots show a simple dropdown, suggesting Google is prioritizing ease of use over complexity.
The timing of this leak is no coincidence. Google I/O 2026 is expected to focus heavily on Gemini’s evolution from a chatbot into a full-fledged digital assistant capable of handling complex, multi-step tasks. The Thinking Level feature fits precisely into that narrative: it gives users the ability to command the AI with varying levels of autonomy and depth. This could be a key differentiator against competitors like ChatGPT and Claude, which currently offer fewer such controls in their consumer apps. While developers have access to reasoning parameters via APIs, typical users are left with binary choices like “use GPT-4 or GPT-3.5.” Google’s approach democratizes that flexibility.
Beyond the Thinking Level, Google also appears to be expanding Gemini’s growing ecosystem of third-party app integrations. Right now, Gemini already works with apps and services like GitHub, OpenStax, Spotify, and WhatsApp. However, support documentation reportedly hints that integrations for Canva, Instacart, and OpenTable are also on the way. None of these integrations appears to be live yet, but the timing makes sense. Google I/O is usually where the company shows off Gemini becoming less of a chatbot and more of a proper digital assistant that can actually do things across apps and services. For instance, imagine asking Gemini to create a presentation on Canva, order groceries via Instacart, and book a dinner reservation on OpenTable — all in one conversation, with the AI handling the logic and sequencing.
These integrations would work through Gemini’s recently launched “Extensions” system, which allows the AI to connect to external services via a unified interface. Each extension handles authentication and data retrieval, while Gemini orchestrates the overall task. The addition of Canva, Instacart, and OpenTable would significantly expand Gemini’s utility beyond information retrieval into action taking. It positions Gemini closer to a virtual assistant like Alexa or Siri, but with far deeper reasoning capabilities — especially when combined with the Thinking Level control.
However, this evolution also raises questions about privacy and data handling. Third-party integrations require sharing user requests with external services, and Google has been cautious about transparency. The company recently published a detailed privacy policy for Gemini Extensions, explaining that data will be used only for fulfilling the immediate request and not for training models. Still, users who enable high thinking levels for tasks involving sensitive information might be wary of the increased processing. Google will need to clearly communicate how reasoning depth affects data retention and security.
Another angle is the impact on subscription tiers. Google currently offers Gemini with a free tier that includes limited access to advanced models. If Thinking Level becomes a premium feature — restricting high reasoning to Gemini Advanced subscribers — it could drive more users toward paid plans. Alternatively, Google might keep low and medium levels free while charging for high level, akin to its current model where “Pro” models are locked behind a subscription. This would mirror the AI Studio pricing, where more compute-intensive reasoning incurs higher costs. The leaked support documents did not clarify pricing, but given Google’s history, monetization is inevitable.
Meanwhile, Google is also pushing the boundaries of AI beyond text. The company recently announced that Project Genie can now use real-world imagery from Google Street View to generate interactive virtual environments. This “world model” technology allows users to walk through AI versions of real places, blending reality with AI-generated styles. While separate from Gemini, it demonstrates Google’s broader investment in creating immersive, reasoning-driven experiences. Similarly, Google wants to reinvent the TV remote with Gemini and pointer controls, turning televisions into interactive AI hubs. These efforts all point to a future where AI does not just answer questions but actively participates in shaping digital experiences.
Despite the excitement, not all changes are welcome. Google’s Gemini might be testing weekly limits, and free users won’t love it. A recent screenshot suggests stricter usage tracking and possible caps on interactions, especially when using heavier AI models. This could be a way to manage server costs as more users adopt the Thinking Level feature — higher reasoning levels consume more compute, and unlimited free access would be unsustainable. Google will need to strike a balance between retaining users and covering infrastructure expenses.
At this point, Gemini’s evolution feels less about smarter answers alone and more about turning the app into something that quietly handles parts of your digital life in the background — ideally without making everything feel unnecessarily complicated. The Thinking Level feature, combined with expanding integrations, represents a step toward that vision. By giving users control over reasoning depth, Google acknowledges that intelligence is not one-size-fits-all. Some days you need a quick ‘yes’ or ‘no’ other days you need a detailed analysis. With the new option, Gemini aims to offer both — and let you choose which you need right now.
As with any rumored feature, plans could change before the official announcement at Google I/O. The limited rollout suggests Google is still testing the waters; feedback from early users will likely shape the final implementation. If successful, Thinking Level could become a standard feature in AI assistants, forcing competitors to follow suit. For now, the news gives us a glimpse of how Google envisions the next generation of AI interaction — one where the user is in control of the machine’s mind.
Source: Digital Trends News