Google's new Thinking level toggle lets Gemini 3 Flash and 3.1 Pro users choose compute intensity, as Canva, Instacart, and OpenTable integrations near launch.
Google is giving Gemini users explicit control over how hard the model thinks, rolling out a "Thinking level" toggle in the app just days before I/O 2026.
The feature surfaced in a limited rollout first spotted by 9to5Google. Inside the model picker, a new Thinking level menu presents two options: Standard and Extended. The toggle appears only with Gemini 3 Flash and Gemini 3.1 Pro; selecting the dedicated Thinking model skips the menu entirely, since that model already runs extended reasoning by default.
That structure mirrors what Google AI Studio gives developers, where thinking intensity is a three-level slider. The consumer version condenses it to a two-position switch.
The competitive context
Giving users a reasoning dial is an acknowledgment that one compute level cannot serve every query equally. A quick lookup does not need the same resources as a multi-step coding problem. Making that trade-off visible to mainstream users also gives Google a differentiation point at a moment when the artificial intelligence industry is racing to make models more capable by default rather than more configurable.
Forbes noted in January that ChatGPT had reached 800 million monthly active users in just three years, one of the fastest consumer technology adoptions on record. By March, CNBC reported that OpenAI was orienting aggressively toward high-compute enterprise use cases, with applications CEO Fidji Simo saying the goal was to turn 900 million weekly active users into heavy users. Google's Extended thinking mode lands squarely on that same battleground.
I/O 2026 is almost certainly the context. Rolling a feature out in limited beta days before a keynote is standard pre-seeding: by the time the announcement lands on stage, some users are already running it, which closes the gap between announcement and credibility.
Third-party integrations
The thinking controls are not the only feature moving. 9to5Google also surfaced support documents showing Canva, Instacart, and OpenTable queued as the next wave of Gemini app integrations. None of these has launched yet.
Canva would let users generate design concepts in Gemini and carry assets directly into production work on the Canva platform. Instacart would accept natural-language grocery requests, use a saved default address to find nearby stores, and drop selected items into a cart. OpenTable would handle restaurant search and booking via Reserve with Google, completing reservations inside the conversation.
The existing third-party roster already includes GitHub, Spotify, WhatsApp, and OpenStax. A pattern emerges: Google is building Gemini as a consumer command surface, not just a developer or enterprise tool. Lifestyle integrations are the near-term priority.
Lock-in through utility is the underlying logic. Each integration that works reliably is a reason a user stays inside Gemini rather than context-switching to a competing assistant. Microsoft has pursued the same strategy with Copilot inside the Office ecosystem; reception has been mixed, partly because the integrations often feel bolted on rather than native. Google has a structural advantage through Android and its own consumer services, positioning those lifestyle integrations closer to where users already spend attention.
The Extended thinking toggle carries weight beyond its immediate function. Reasoning-capable artificial intelligence has been a developer-facing concept for most of its short history, largely because it is slow and expensive. Putting a thinking intensity control in a consumer app normalizes the category: users learn that the output they receive reflects how much computation was committed to the query, not a fixed ceiling of the model.
If I/O 2026 unveils a new Gemini model with competitive pricing and extended thinking at scale, this week's rollout will read as prologue. If the keynote is incremental, the toggle may be the main event after all.
FAQ
What is Gemini's Extended thinking level?
Extended thinking instructs the model to reason through a query more thoroughly before generating a response. Users can choose between Standard and Extended in a new Thinking level menu in the Gemini app.
Which Gemini models support the Thinking level toggle?
The setting appears when selecting Gemini 3 Flash or Gemini 3.1 Pro. It does not show up with the dedicated Thinking model, which already applies extended reasoning by default.
What new third-party apps is Gemini adding?
Canva, Instacart, and OpenTable appear in support documents as upcoming integrations. Current connections include GitHub, Spotify, WhatsApp, and OpenStax.
Is this connected to Google I/O 2026?
Almost certainly. The timing of the limited rollout follows a pattern Google has used before to prime features ahead of a major keynote announcement.