My Fresh LLM Benchmark and Plan-Code-Review Methodology
AI Coding news: June 24, 2026.
Hey hey,
This week, I spent a lot of time on re-testing LLMs for my benchmark and found a few important “twists” to make the numbers more accurate, you will see a few videos about it below.
Also, my second course on AI Coding Daily: I’m showing my methodology how I build the realistic project trying to save time/money using the *mix* of LLMs.
Finally, a few pieces of news from the AI coding community, but it seems like there are fewer loud announcements and people are more in the “deep work” mode trying to actually use the existing LLMs more efficiently.
For Premium Members
[NEW Course] Plan-Code-Review: Building a Real Upwork Project in 1 Hour
In this 45-minute course, I will show how I build a real project from Upwork job description, using Opus for planning, and other models for implementation and code review.
My YouTube videos
I Found One Prompt Change For Better AI Code
I think I discovered the main reason why frontier LLMs score 20/20 on my benchmark and cheaper models do not, let me show you.
My NEW LLM Coding Score: Models Often Fail at THIS
I’ve benchmarked the models on one new project: found the aspect where only frontier models scored higher points, and none of the models scored perfect 5 out of 5.
I Tested FREE Nvidia Nemotron Model on 5 Coding Tasks
Before this video, I hadn’t heard about a free LLM like Nemotron, let me show you how much it scored on my traditional benchmark.
[Text Post] You asked in the comments to test GPT-5.4-Mini on my LLM benchmark. I did. Wow.
Almost same result on daily coding tasks, but 6x cheaper than GPT-5.5 Medium.
AI Coding Community News
Introducing Claude Tag, a new way for teams to work with Claude. — x.com
In Slack, Claude joins as a team member with access to the channels and tools you choose. Tag Claude in and delegate tasks to it while you focus on other work.
Claude Code now supports artifacts — claude.com
Preview your in-progress work in Claude Code as a live, interactive artifact—built from your full session context and shareable with your team.
multica-ai/andrej-karpathy-skills — github.com
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy’s observations on LLM coding pitfalls.
rtk-ai/rtk — github.com
CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies.
Cursor: We’re launching code storage and git hosting — x.com
Origin gives teams and agents a place to host, review, and collaborate on code. Available this fall. Join the waitlist.
That’s a wrap for this week. Keep building with AI!
Povilas Korop
AICodingDaily.com