Gemini 3 Pro gives Cline more headroom for long-running coding

Gemini 3 Pro is Google’s new flagship model for reasoning, multimodal work, and coding. On public leaderboards it lands where you would expect a frontier model to land: top scores on web dev and reasoning benchmarks, strong results on terminal and tool-use evaluations, and a clear step up from Gemini 2.5 Pro.

Benchmarks are not the whole story, but they do tell us something useful here. Gemini 3 Pro is comfortable operating over a codebase, tools, and artifacts, not just answering isolated questions. That is the territory Cline cares about.

How it feels in Cline

In Cline, the difference with Gemini 3 Pro shows up as you push the frontier on model capabilities. A few patterns are already clear:

It holds onto structure over longer tasks. When you ask it to implement a feature that spans multiple files and layers, it is better at sticking to the original plan instead of wandering into unrelated edits halfway through.

It makes better use of long context. The 1M-token window is not just a number; the model actually uses that extra context to respect existing patterns, touch related files together, and keep changes coherent with the rest of the project. The effective context feels larger than 2.5 Pro's 1M token context window. We highly recommend letting Gemini 3 Pro dig deep during plan mode and utilizing /deep-planning.

Its ability to generate quality frontend code with good taste is impressive. Our internal testing has shown a significant leap forward from 2.5 Pro and competitiveness with Sonnet 4.5, which has widely been regarded as the leading model for generating UI, especially with 3D, game-like creations.

Its multimodal understanding translates cleanly into everyday work. Mixing code, design screenshots, and documentation in one conversation is more productive when the model can tie them together into concrete implementation steps instead of treating them as separate prompts.

For Cline users, the net effect is simple: more of the tasks that used to feel too big to hand off now fall into the range where you can let the agent drive while you review.

How it behaves in Cline as an agent

Cline already pushes models into a structured workflow: investigate the project, outline the work, then execute in small, reviewable steps.

Gemini 3 Pro slots into that pattern without special handling. It reads broadly enough to understand how your project is wired together (with a context window large enough to hold much of your project's context), produces thorough plans that reference real files and functions, and then works through those plans in a way that keeps earlier context in view.

Gemini 3 Pro raises the ceiling of what you can accomplish using Cline.

Getting started

Gemini 3 Pro is available in Cline through the Cline, OpenRouter, Vercel, Google Gemini, and GCP Vertex providers.

If you end up with interesting successes, failures, or patterns:

Reddit: https://www.reddit.com/r/cline/
Discord: https://discord.gg/cline
Docs: https://docs.cline.bot

We will keep tuning Cline’s defaults and prompts as we see more real usage with Gemini 3 Pro, but even at launch it already expands what feels reasonable to hand off to an agent in your own editor.