13 Seconds and 1.8 Cents: The Era of Disposable Code

How much should it cost to bring a spontaneous digital idea to life? If you are using Google’s Gemini 3.8 Flash model, the answer might be as little as 1.8 cents.

While the artificial intelligence industry often fixates on massive, computationally heavy "frontier" models that can pass bar exams or write novels, a quieter revolution is happening at the other end of the spectrum. The focus is shifting toward models optimized for blistering speed and micro-transactional costs. A recent experiment by developer Simon Willison perfectly illustrates why this shift matters to anyone who uses the internet.

Willison decided to test the Gemini 3.8 Flash model with a delightfully vague prompt: "make me a cool thing in html." He didn't provide detailed technical specifications or complex architectural guidelines. Despite the lack of direction, the AI delivered a fully functional, interactive piece of HTML and JavaScript. The most impressive part wasn't just the output, but the metrics behind it: the entire generation process took a mere 13 seconds and cost less than two cents.

This iteration of Gemini introduces a fascinating feature to help manage these costs and speeds: adjustable "thinking levels." Users can dial the AI's cognitive effort to low, medium, or high depending on the task at hand. (Google also released a specialized "Cyber" version of the model, though it remains restricted to trusted cybersecurity defenders). By allowing users to scale the AI's effort, the technology becomes incredibly versatile, adapting to whether you need a quick script or a more deeply reasoned solution.

To prove its utility beyond simple party tricks, Willison also deployed Gemini 3.8 Flash to do actual software maintenance. Using a coding agent plugin, he instructed the model to upgrade his personal Markdown rendering tool. The AI successfully added a feature that allows HTML blocks to be safely rendered within sandboxed iframes, proving it can handle practical, everyday coding chores.

The broader implication of this technology extends far beyond making life easier for professional programmers. We are entering an era of "disposable software," where the barrier to creating functional web components is practically non-existent. When generating code is this fast and cheap, experimentation carries zero risk. You no longer need to know how to write JavaScript to build a custom tool for your daily workflow; you just need to know what you want to build. As AI continues to commoditize code generation, the most valuable skill will simply be the curiosity to ask it to build something cool.

Key Points

• Google's Gemini 3.8 Flash model is optimized for high speed and extremely low generation costs.
• The model features adjustable 'thinking levels' to scale computational effort based on task complexity.
• In a real-world test, it generated functional HTML and JavaScript in just 13 seconds for 1.8 cents.
• Ultra-cheap AI coding capabilities dramatically lower the barrier to entry for everyday digital experimentation.

Why It Matters

By reducing the cost of coding to mere pennies, AI models are transforming software creation from a specialized profession into an accessible, everyday tool for everyone.

Sources:

• llm-gemini 0.34 — Simon Willison's Weblog