Gemini 3 Pro Preview is Now Supported in Kilo Code

Here's what you need to know.

Google has just released Gemini 3, their most advanced model yet, bringing state-of-the-art reasoning, richer multimodal understanding, and dramatically more capable agentic behavior.

Gemini 3 Pro Preview is available today across the entire Kilo Code ecosystem. You can use it through:

• The Kilo Gateway in the VSCode/JetBrains Extension

  • Your first $10 top-up still gets you +$20 bonus credits

• Your own Google API key

Just type “Gemini 3” in the model selector inside Kilo Code, and you’ll see it immediately:

---

Gemini 3 — What’s New

Gemini 3 represents a major leap over the 1.x and 2.x families. Google describes it as their most intelligent model to date, built around deeper reasoning, more reliable problem-solving, and far more fluent multimodal understanding.

Here are the benchmarks:

Reasoning Performance

Gemini 3 tops the LMArena leaderboard with a 1501 Elo score and delivers frontier-level performance across multiple reasoning benchmarks:

• GPQA Diamond: 91.9%

• Humanity’s Last Exam: 37.5% (no tools)

• MathArena Apex: 23.4%

This means better depth, nuance, and consistency on complex tasks like architecture planning, debugging multi-file systems, and scientific problem-solving.

---

Coding With Gemini 3

Gemini 3 is built for richer, more interactive coding experiences — which are now fully available in the Kilo Code extension and CLI.

In our own tests, Gemini 3 Pro Preview outperformed other top models on five challenging UI and application-building tasks. Across scenarios like building interactive physics sandboxes, designing workflow editors, or creating browser-based micro-OS environments, Gemini 3 consistently produced cleaner, more intentional solutions:

What stood out most was how Gemini writes code:

• It uses modern, sensible libraries without over-engineering

• Layouts and UIs adapt intelligently to constraints rather than feeling template-driven

• It reliably selects correct asset/CDN paths and exhibits strong awareness of current tooling

These qualitative differences matter in real development workflows, where correctness, structure, and maintainability often outweigh raw token-level metrics.

---

Planning, Agents, and Autonomy

Gemini 3 continues Google’s push into agentic workflows. It now leads long-horizon planning benchmarks:

• Vending-Bench 2: best-in-class returns for long-horizon simulations

• More consistent tool use

• Better stepwise planning

• More stable long-running sessions

In real workflows, this means:

• More reliable multi-step refactors

• Better repository-wide changes

• More accurate filesystem navigation

• More predictable tool usage inside Kilo Code

It’s compatible with Kilo’s multi-model workflows, meaning you can do intensive reasoning and architecture tasks with Gemini 3.0, and switch to a more cost-efficient model like Gemini 2.5 Flash Lite for minor fixes.

---

Pricing

Input Pricing

• Up to 200K tokens: $2 per 1M tokens

• Above 200K tokens: $4 per 1M tokens

Output Pricing

• Up to 200K tokens: $12 per 1M tokens

• Above 200K tokens: $18 per 1M tokens

Cache Read

• Up to 200K tokens: $0.20 per 1M tokens

• Above 200K tokens: $0.40 per 1M tokens

Cache Write

• Up to 200K tokens: $2.375 per 1M tokens

• Above 200K tokens: $4.375 per 1M tokens

Total Context: 1.05M tokens
Max Output: 65.5K tokens

Live Q&A Event

On November 19th, 2025 at 12:00 PM EST, we’ll host a live Q&A webinar where you’ll learn what sets this Google model apart from others.

10 people who join the webinar will get $100 in LLM usage credits via our Kilo Code Gateway. You’ll learn whether Gemini 3 Pro Preview is worth the hype just 24 hours after its release, and you can ask questions and discuss the model with members of the Kilo Code team.