How This Simple Fix Transforms Gemini 3.1 Pro with GLM-5?

I made a video about Gemini 3.1 Pro recently and I was pretty harsh on it. It regressed on my one-shot benchmark and dropped massively on agentic tasks. It plans forever, wastes tokens, and doesn't use tools properly.

I basically said don't pay for this model. A lot of you said I should try it with the King Mode prompt. After testing it, yeah, it actually works.

Gemini 3.1 Pro is fixable and King Mode is the fix. King Mode is a system prompt I built to make models stop being lazy. I’ve used it on GLM 4.7 and GLM-5, and the results are always the same.

It doesn't make models smarter. It makes them more disciplined. It strips the fluff, forces focus, and turns inconsistent models into consistent ones.

Gemini 3.1 Pro’s problems are discipline problems, not intelligence problems. The model is actually smart with a 1 million token context window. It scores 77.1% on ARC AGI2, and Google’s benchmarks show 80.6 on SWEBench.

It just doesn't know when to stop thinking and start doing.

Why this simple fix transforms Gemini 3.1 Pro with GLM-5

Remember the issues from my last review. It plans for 90 seconds before writing a single line of code. It repeats itself in its thinking with things like "contemplating the design" then "mapping the layout" then "planning the implementation" which are all the same thing.

It's burning tokens on nothing. King Mode’s Ultraink trigger fixes this. It tells the model to assess the complexity, decide if it needs deep reasoning or quick execution, and commit.

The zero fluff directive kills the repetitive planning. Instead of 90 seconds of planning, you get 10 seconds and then it starts building. That alone makes the model way more usable.

Discipline over intelligence in How This Simple Fix Transforms Gemini 3.1 Pro with GLM-5?

I tested it on the SvelteKit Kanban task without King Mode. It did 90 seconds of planning across two phases. The second plan was basically a copy of the first.

With King Mode, it did one focused planning phase of about 15 seconds and then went straight to coding. The output was better too because it wasn’t wasting its context on redundant thinking. That single change flips the experience from frustrating to productive.

Read More: Fix Google Chrome Opening New Tabs

Front end strengths in How This Simple Fix Transforms Gemini 3.1 Pro with GLM-5

Gemini 3.1 Pro with King Mode is actually really good at front end. When you strip away the planning bloat and let the model focus on UI work, it's genuinely impressive. The Tailwind implementations are clean and the component structures are well organized.

It understands responsive design really well. With that massive context window, it can hold an entire design system in memory while building components. I think Google’s training data just had a ton of front-end code and documentation, and you can feel it in the output.

Give it a detailed UI prompt and it gives you polished, production-looking components. Not just functional code, actually good-looking code. That’s the difference when King Mode keeps it focused.

Verdant setup for How This Simple Fix Transforms Gemini 3.1 Pro with GLM-5

I have Verdant open with the model set to Gemini 3.1 Pro. I’ve loaded the King Mode prompt into the project rules. I’m going to give it a simple front-end task.

I type Ultraink followed by the task. Here’s the exact prompt I use:

Ultraink. Build a modern portfolio website using Next.js 14 with Tailwind. Create a hero section with animated text, a projects grid with hover effects, and a contact form. Use a dark theme with purple accent colors. Make it look premium and clean.

Because of King Mode, Gemini doesn’t go into a 90-second planning loop. It gets a quick focused plan and starts building immediately.

What it builds under How This Simple Fix Transforms Gemini 3.1 Pro with GLM-5

It outputs components, sets up the layout, and adds animations. The hero section has smooth text transitions, the project cards have a nice hover scale effect with shadows, and the contact form is clean with proper validation states. The dark theme with purple accents looks really good, like it has design taste.

For front end work, Gemini 3.1 Pro with King Mode is genuinely strong. The visual output quality is there and it’s fast because King Mode stops it from wasting time on long planning. That combo makes it reliable for UI builds.

Read More: Fix Google Chrome Untitled Blank Page

Back end gaps and pairing in How This Simple Fix Transforms Gemini 3.1 Pro with GLM-5

The back end is a different story. Complex architecture, database design, API logic, and security policies are where Gemini still falls apart. It doesn’t have that architectural intuition.

You don’t need to use one model for everything. GLM-5 with King Mode is an absolute beast at backend work. It’s a 744 billion parameter model that Z AI built to be a system architect.

It’s number one on my agentic leaderboard and it beat Opus 4.6 on basically everything. It’s free to use in Kilo Code. In Verdant, you can easily run both.

Spin up one agent with GLM-5 for the back end. Spin up another agent with Gemini 3.1 Pro for the front end. Verdant handles the parallel execution with isolated git work trees so they don’t conflict.

Merge the work trees when they’re done. It’s that simple. GLM-5 handles the hard architectural stuff, Gemini handles the visual stuff, and Verdant orchestrates the whole thing.

The total cost is basically nothing because both models are either free or dirt cheap. It’s like having a senior back-end architect and a talented front-end developer working in parallel. Except your team works for free.

Step-by-step in Verdant for How This Simple Fix Transforms Gemini 3.1 Pro with GLM-5

Inject the King Mode prompt into your Verdant project rules so it applies to all agents.

Create an agent with GLM-5 scoped to backend responsibilities like architecture, database, APIs, and policies.

Create a second agent with Gemini 3.1 Pro scoped to front-end responsibilities like components, layouts, and styles.

Enable parallel execution with isolated git work trees to prevent conflicts during build.

When both agents finish, merge the work trees and resolve any minor integration details.

Read More: Fix Google Chrome White Screen Windows

Tips that stick in How This Simple Fix Transforms Gemini 3.1 Pro with GLM-5

Inject the King Mode prompt into your Verdant project rules so it consistently applies to both models. Both GLM-5 and Gemini benefit from it, but Gemini benefits more because its discipline issues are worse. That single change cuts planning loops and waste.

Use the Ultraink prefix for anything complex. For simple stuff like changing a button color, you don’t need it. For multi-component work, it makes a big difference.

Keep your front-end prompts descriptive. Gemini responds really well to visual direction. Don’t just say, "Build a dashboard," say, "Build a premium dark themed dashboard with blue accents and smooth transitions."

Final thoughts on How This Simple Fix Transforms Gemini 3.1 Pro with GLM-5

King Mode fixes Gemini 3.1 Pro’s discipline, not its intelligence. With Ultraink and zero fluff, it shifts from over-planning to focused execution. Paired with GLM-5 for back end and Verdant for orchestration, you get fast, clean front end, solid architecture, and almost no cost.