With Google I/O 2026 on the horizon, the company appears to be accelerating its push to refine Gemini, its flagship AI assistant. A recent leak, spotted by 9to5Google, reveals a new feature called "Thinking Level" inside the Gemini app. This option, currently visible to a limited set of users, allows individuals to dial up or down the amount of cognitive effort Gemini exerts before generating a response. The move aligns with broader industry trends where AI companies are competing to offer more controlled, agentic, and thoughtful interactions.

The Thinking Level feature resides in Gemini’s existing model picker, which already offers choices like Fast, Thinking, Pro, or Google AI Plus. Now, when selecting the Fast (Gemini 3 Flash) or Gemini 3.1 Pro models with thinking enabled, users see an additional slider or dropdown to adjust reasoning depth. While the exact levels are not yet fully documented, early reports suggest options mirror those in Google AI Studio: Low, Medium, and High. This granularity empowers users to trade off between speed and thoroughness, depending on the task at hand.

Why reasoning depth matters

Not every query requires the full might of a AI’s reasoning engine. Asking for a quick fact, a recipe, or a simple calculation often benefits from low-latency responses. Conversely, complex analysis, creative writing, or problem-solving may justify additional processing time. By giving users explicit control, Google is effectively acknowledging that one-size-fits-all AI responses can be either wasteful or insufficient. This is a subtle but significant shift from the black-box approach many chatbots employ today.

Competitors are also moving in this direction. OpenAI recently introduced reasoning models like o1, which allow users to specify “high” effort for deeper reasoning. Anthropic’s Claude sometimes provides thinking traces. But Google’s integration of a sliding scale directly into the consumer app could set a new standard for user empowerment. The level of customization potentially improves user satisfaction and trust, as people can understand and influence how the AI arrives at its conclusions.

From AI Studio to the real world

Google AI Studio, the company’s developer-focused platform, has long offered Low, Medium, and High reasoning settings. Bringing this to the mainstream Gemini app bridges the gap between professional developers and everyday users. It suggests Google views AI flexibility as a core product feature, not just an experimental tool. The company has been iterating rapidly on Gemini since its public launch, with updates ranging from multimodal capabilities to integration with services like GitHub, Spotify, and WhatsApp.

These developments come at a time when AI assistants are evolving from simple chatbots into proactive digital companions. Google recently hinted at deeper third-party app integrations, including Canva, Instacart, and OpenTable. While none are live yet, the timing suggests Google I/O will showcase Gemini as less of a standalone tool and more of an orchestration layer for daily life. The Thinking Level feature fits perfectly into that vision, allowing Gemini to adapt its cognitive load to the context—quick for routine actions, thorough for complex tasks.

Broader implications for AI usability

The psychological impact of giving users control over AI reasoning should not be underestimated. Studies in human-computer interaction show that perceived control increases trust and satisfaction. By letting users choose whether Gemini thinks “harder” or “softer,” Google empowers them to align the AI’s behavior with their immediate needs. This could reduce frustration from overthinking trivial requests or underthinking critical ones.

Moreover, such granularity could help differentiate Gemini in a crowded market. ChatGPT, Copilot, and other assistants often rely on auto-tuning or preset modes like “Creative” versus “Precise.” A direct reasoning slider is more transparent and might appeal to power users who want to fine-tune performance. It also opens the door to potential monetization tiers—perhaps faster thinking for free users, deeper reasoning for subscribers—though Google has not announced such plans.

Historical context: Google’s AI journey

Google’s AI efforts span decades, from early research in natural language processing to the launch of Google Assistant in 2016. Gemini represents the next generation, built on a massive multimodal model. The Thinking Level feature echoes the company’s philosophy of making AI more accessible and controllable. Back in 2023, Google introduced Bard with limited capabilities; user feedback demanded more nuance and customization. Since then, the team has pivoted to iterative, user-driven improvements.

The feature’s appearance just before I/O is strategic. Google traditionally uses its developer conference to unveil major Gemini updates. Last year, the focus was on multimodal and agentic capabilities. This year, the emphasis seems to be on fine-grained control and depth. Combined with the rumored expansion of app integrations, Gemini is evolving into a platform that can handle both trivial and complex requests without overwhelming users.

Industry analysts note that AI reasoning depth is becoming a key battleground. Microsoft’s Copilot, for instance, offers different creativity levels. Apple Intelligence uses on-device processing for privacy but with limited depth. Google’s cloud-based approach with adjustable reasoning could give it an edge in scenarios requiring both speed and accuracy. For example, a user might want a quick answer to “What’s the weather?” (low thinking) versus “explain the implications of quantum computing on cryptography” (high thinking).

Another angle is the environmental cost. Deeper reasoning consumes more computational resources and energy. By letting users choose, Google may also be encouraging responsible AI usage—running lightweight tasks on simpler models and reserving heavy lifting for when it’s truly needed. This aligns with corporate sustainability goals and could appeal to eco-conscious consumers.

In practice, the Thinking Level slider could appear as a small icon next to the model name. Tapping it might reveal three to five levels, from “Fast” to “Deep.” Each level would adjust parameters such as token generation time, chain-of-thought processing, and context window usage. Google would likely provide tooltips explaining the trade-offs, helping users make informed choices.

Early testers have reported that the feature is stable but still rough around the edges. Some mention that higher thinking levels sometimes produce overly verbose answers, even for simple queries. This suggests that fine-tuning is ongoing. However, the feedback loop from limited rollout should help Google refine the experience before a broader launch.

Looking ahead, the Thinking Level feature could become a staple of not just Gemini but also Google’s other AI products, such as Workspace integration or Google Cloud AI. The ability to control reasoning depth might extend to agents, giving developers and users the same flexibility. This is a logical step toward a future where AI is not a monolithic black box but a customizable tool that adapts to its operator’s preferences.

As Google I/O 2026 approaches, the industry watches closely. The Thinking Level leak is just one of several hints that Google is ready to double down on user-centric AI design. If executed well, it could redefine how people interact with AI assistants—not as passive answer machines, but as active partners with adjustable intellect. For now, the feature remains behind a limited test, but its potential impact is clear: more control, more trust, and more practical utility in daily digital life.

Source: Digital Trends News