Originally published as a Advisory Hour Substack post on 2026-07-13. This first-party site copy preserves the text, localizes public supporting media, and connects the record to its project trail.
Original post
I posted a brief note on Thursday evening that putting 5.6 Sol on tasks that 5.5 hadn’t yet solved brought substantial quality jumps for no more work from me than “pick the smarter model”. OpenAI might be the first company in history to realize the true promise and value of “just update the config and it gets better”.
The note I posted (which I have the photo from it at the end of this article) got my thinking that I should stretch to find the limit. Eventually I formed a question. The question I had in mind was “Can 5.6 make a documentary? A two minute flick?”
Incredibly, it could. I was shocked when I heard a narrator. That was just a “eh, maybe it’ll solve that too-ha yeah, right.” And then it did.
The stack
ChatGPT-5.6 Sol
Hermes
Blender MCP
Hermes solved the narrator by using Microsoft Edge TTS on ubuntu, which is one of the default providers out of Hermes. I didn’t even remember setting it up and had to ask Hermes about that one. The SFX were selected by Hermes from a SFX collection that I’ve collected over the years for different projects.
I’m not really sure what to say. I’ve reviewed different scenes and such. It’s not going to win any awards for polish or sizzle-but it exists.
The entire movie, art direction, cinematography and everything was selected by Hermes driving with Sol and the tooling it had access to in order to build that video. It maintained some proof stills, modeled the sun, a passing version of the earth, robots, satellites and more.
There’s clearly room for improvement. The sun magnetic flares are fun noodles and the earth is a version of the sun, but blue. I mean, technically that does work. I give the model a bit of a pass here. It constructed the whole thing in one pass.
These build sites (or proof sites, I tend to swap the term around and haven’t quite landed on one) have far more content than I could ever hope to bring over to Substack and frankly I’m not sure I even could, or that the internet really needs this much stuff. Still, it’s fun to see.
I now feel compelled to go thru other projects and just try to uplift them up another level. What happens if you stall out, and then just wait for a smarter ai? When has that been a viable strategy? Well, 5.5 came out in April and then 5.6 comes out in July. We’re in some strange project escape velocity calcuation.
The image on the left was 5.5’s attempt at this. The prompt was to build a early version of the Dyson sphere. My thinking is that I’m likely (as are you) to see one of these now in our lifetimes, so I wanted to see just what that might look like.
