Ritorna Vinci...?
An update - what we've been up to
Forgive the Verdi.
What’s been Going On
It’s been a hot minute. I won’t lie; the value proposition of Verdion evolves faster than a website can keep up. In October of 2025, an adversarial collaboration among a few frontier cloud LLMs was interesting, and it resulted in some novel results (as the MVP thesis shows). There’s still a “there” there.
Since then, AI’s popularity has exploded. There is simply no other word. From the RAMpocalypse to the competing releases of GPT5 (Luna/Terra/Sol) and Claude (Mythos, Fable, Opus 5), to the rise of very worthwhile locally-hostable agentic LLMs (Gemma-4 31B and Qwen3.6 27B Dense/35B A3B, both more than useable on a 24GB RTX at 256k context) - the state of the art changes hourly.
Verdion
Where does Verdion fit? Well, everywhere, from where I stand. It’s a plug-and-play reasoning evaluation platform as easily as it is an adversarial collaboration reasoning engine. Modularity and portability was a major focus while I built it, and it shows; moving it from talking to Claude, GPT, and Gemini to comparing output from Qwen, Llama, and Gemma was a configuration file change.
Verdion has more than paid for itself in my own meta-workflow. I think it can be helpful for others’ too.
Where we are
So the story hasn’t changed, but it’s broadened considerably. The case for Verdion has become all the stronger. Things are just becoming more and more interesting.
Keep your eyes peeled!