Jules Leaves Beta With Tiered Limits and Gemini 2.5 Pro Planning

What actually shipped

Jules is now public and powered by Gemini 2.5. Google says that during the beta, thousands of developers ran tens of thousands of tasks, producing over 140,000 code improvements shared publicly. That is the concrete usage record Google is citing to justify the graduation from beta.

The changes are incremental and driven by feedback: a polished interface, hundreds of bug fixes, and new capabilities. Three are named specifically — reusing previous setups so new tasks start faster, GitHub issues integration, and multimodal support. These are workflow features, not model rewrites.

The planning step is the substantive change

The most technically meaningful line in the announcement is that Jules now uses the advanced thinking capabilities of Gemini 2.5 Pro to develop coding plans. Google frames this as producing higher-quality code outputs.

This is a plan-first architecture: the agent reasons about an approach before emitting code, rather than generating directly. For an asynchronous agent that runs tasks without a developer watching, front-loading a planning stage is a defensible design choice — the plan becomes the artifact a developer can inspect and correct when they return.

Access is now a capacity ladder

The pricing structure is the real story for teams deciding how to use Jules. Google introduced three structured tiers: introductory access for trying Jules across projects, Jules in Google AI Pro at 5x higher limits for daily coding, and Jules in Google AI Ultra at 20x higher limits, which Google describes as built for intensive, multi-agent workflows at scale.

Notably, the differentiator between tiers is usage limits, not distinct features. The same agent runs at every level; you pay for how much you can run. Google does not publish the actual numbers in the post, deferring them to jules.google, so the 5x and 20x multipliers are relative to an introductory baseline that is not stated here.

The rollout starts today for AI Pro and Ultra subscribers, and eligible college students can sign up for a free year of AI Pro — a distribution move that seeds the tool with students before they reach paying teams.

What the tier design implies for teams planning agent budgets

Because Jules gates capacity rather than capability, adoption planning becomes a throughput question. A team evaluating Jules should model how many asynchronous tasks it expects to dispatch per developer per day, then map that to the Pro or Ultra limits rather than shopping for feature parity.

The explicit call-out of multi-agent workflows at scale on the Ultra tier signals where Google expects heavy usage to concentrate: fleets of agents running concurrent tasks. Teams that intend to fan out work across many parallel Jules runs will hit limits fastest, and the published caps at jules.google — not the marketing multipliers — are what should drive the decision.