Frontier AI at half the cost: Gemini 3.1 Pro now available in Augment Code

Starting today, Gemini 3.1 Pro is available as a model option in Augment.

We ran it head-to-head against Opus 4.6 on the same task: plan out adding a new field end-to-end across the codebase. Gemini caught a subtlety Opus missed. That got our attention.

Gemini is good at this kind of structural thinking. Even more impressive: it ran at 2.6x cheaper per message in real usage compared to Opus 4.6. For engineers using Augment heavily, that savings compounds.

It's not every day that a new model makes it into Augment Code. We want to balance between giving professional engineers choice, with a guarantee of quality and performance. Gemini 3.1 Pro is here with GPT 5.4 and Claude 4.6 because it earned it.

How Gemini compares

*Based on a standard medium-complexity task. Actual credit consumption varies based on task complexity, context size, and response length.

The quality gap between Gemini and Opus in day-to-day work is smaller than the price gap suggests. For most changes, implementing a feature, tracking down a bug, navigating unfamiliar code, the output is often equivalent or identical to what Opus produces.

Where it earns its place

Gemini is strongest at reasoning through changes before making them. That's where the head-to-head result came from: given a task that required thinking through cascading effects across the codebase, Gemini's plan was more thorough.

It's also good for debugging and investigation. A workflow that works well: ask Gemini to orient itself and summarize what's relevant before you start making changes. It surfaces the right context faster and gets you to the fix sooner.

For everyday execution, most changes come out at the same level as Opus. Planning first makes the execution cleaner, but even without it, Gemini produces equivalent output on targeted, well-scoped tasks.

What to know before you try Gemini 3.1 Pro

Gemini is slightly less good at inferring intent than Opus. Opus often picks up on what you mean even when you're vague about next steps. With Gemini, being explicit helps: if you want it to make changes, say so. If you want it to investigate and write a plan before touching anything, say that too.

It can also lose the thread in long conversations or when tool outputs get large. The fix is the same approach that gets the best out of it anyway: break work into chunks, plan before executing. If a task is getting complex, stop and reorient before continuing.

If you want both of those problems solved by default, that's what Intent is for. Intent is Augment's agent orchestration workspace, built around spec-driven development. A coordinator breaks work into discrete tasks before anything executes — so the planning-first workflow happens automatically, and scope stays tight throughout.

Getting started

Select Gemini 3.1 Pro from the model picker in Augment. The workflow that tends to work best: planning prompt first, then execution. Ask it to think through what needs to change before it touches anything. That's where it catches things it would otherwise miss, and it makes the execution steps cleaner.

At 2.6x cheaper per message, the savings are real for engineers who use Augment all day. For investigative and planning tasks especially, try Gemini as your first pass. Most of the time, you won't need to switch.