Keeping token cost down and evaluation test

I asked tree different LLMs and AI "coder appa" to writhe a plan for a Tesla Charging app, with the goal of not using the token limit agains the Tesla APi (10$) free pr month. and to make the none High-end LLMs to make the plan. and i used /grill-me for all of them

Jan Petersen

July 4, 2026

1 min read

14 views

The test goal, was to get a Charging app Plan, from each LLM and make them evaluted the tree plan, choice a Winner and the combine a "Winner-plan" from from all the plans.

**The tools and LLM's:

Open code: Big Pickle

VS code: Kimi K2.7

Github App: Mai-Coder-1-Flash

the start prompt for all of them

/grill-me if i like to build a app thats primary function is to control changing on the cheapest hours of the day, but i like it to stay in the APi limits $10 is provided to support individual developers/small applications.

more information on limits here: https://developer.tesla.com/docs/fleet-api/billing-and-limits?utm_source=copilot.com#rate-limits

the app should be able to control and plan to charge the tesla car, but not pull the API for information so we use the 10$ we have, so /goal is the car should beready in the morgning chaged the % we have asked it to do, and get the power price from the internet its in Denmark

Here are the evaluation from each: https://jp-my-blog.vercel.app/blog/big-pickle-plan

https://jp-my-blog.vercel.app/blog/plan-kimi-k27-coder

https://jp-my-blog.vercel.app/blog/plan-mai

and this is there own plans Each LLM's own plan

https://jp-my-blog.vercel.app/blog/plan-kimi-k27-coder-own-plan

https://jp-my-blog.vercel.app/blog/plan-big-pickle-own-plan

https://jp-my-blog.vercel.app/blog/plan-mai-own-plan