GPT-6.1 Sol is HERE!
AI News Today | Julian Goldie Podcast
September 30, 2026
GPT-6.1 Sol is HERE!
Speakers Julian Goldie
Julian Goldie (0:00)
GPT-61 Sol just dropped, and we're going to run through some of the cool stuff we've built with it, how it works, how it performs versus other models, and also what the difference is between GPT-61 Sol and GPT-6 Astra. And I'll showcase some of the cool stuff that we've actually built with it, because it can build some pretty amazing stuff, which we'll come on to later, all right? So let's talk about this. Basically, GPT-6 Sol, if you're wondering what's the difference, it nearly matches OpenAI's top model at a fifth of the price. So I actually gave it one prompt per word in Codex, and it built like full 3D games. And we're going to walk through exactly how it works. Now, you can get access to it inside Codex. If you're inside ChatGPT, for example, you can go inside this section. And then we can switch to GPT-61, as you can see right here. Now, we can also switch all the way up to Ultra.
And if we lower down the settings, then we've also got light as an option too. And we can change the models as well, as you can see right here. So you still have Astra, 5.6 Sol, 6 Sol, but 6.1 Sol is the newest version. So what are the differences here? Well, if we actually look at the price per million output tokens, GPT-6 Astra is 50, right? Whereas for example, GPT-61 Sol is 10 So this is a fifth of GPT-6 Astra's token price. And it's $2 per million input, $10 per million output, $4.10 per million cached input tokens. For everything that we built, it took about 30 minutes for the average build. And we had 129,000 tokens used per word. And also it's much faster. So according to the research, eight times faster, ultrafast mode is also coming to Codex soon too. So there will be an ultrafast version of GPT-61 sole as well. So what's actually better here and what the difference is? So this is pretty much matching Astra on the benchmarks.
It's cheaper, it's better reducing hallucinations of factual errors.
And you can see how it performs on Deep SWE over here.
So on Deep SWE, it's pretty much matching GPT-6 Astra. It's a bigger step up from GPT-6 sole as well. So if we look, for example, at factual errors on hard prompts with low effort, you can see here the GPT-6 sole scores 11.4% versus GPT-61 sole, which scores 7.1%.
So it makes about 32% fewer errors when you're using this.
And it created some awesome stuff as well. So we tested out some 3D builds. Let's open this up and see how it performs. So this is like a 3D world that we built here. I will say the design and the outputs here don't feel as good as something like Sonnet 5.5 or Opus 5.5, but still it can build some cool stuff. So it can build 3D worlds. By the way, if you want to get the prompts from everything that we've done today, I will post that inside the AI Profile Boarding Community. Link in the comments description or go to the AI Profile boarding.com to get access to all of our prompts and our full guide on how this works and also share this guide as well so you can find it. Here's another example. So this is called Sky Harbor. And this is like a fully playable 3D world, so we can open this up, which is not bad again. It's not as good, I would say, as GP6 Astro, but the details here are pretty nice. It builds on like cool 3D builds. Bear in mind, this was done with one prompt as well. So it's pretty impressive for a one prompt system. Here's another example. So this is Deep Blue, which is like a 3D world too, fully playable, underwater.
One thing I'm noticing is the design is just not that great, but it's not bad. It's kind of passable, if that makes sense.
And then a tiny world here as well. Now at this point, you might be going, okay, which one is better? Because there's been two big releases this week for models, right? There's been GPT 6.1 sole and Sonic 5.5. So how do they perform? So this is the outputs from Sonic 5.5. And I would just say it's a much more impressive model overall, right? So if we open this up, this is the 3D world that's been built with Sonic 5.5. And like the details, the quality, the design, everything just looks a lot nicer with it, right? Even the colors and everything else. Here's another example. So we can actually pull up the 3D world that we created with sole versus the world that we've created with Sonic 5.5. So if we open this up, here we go.
6 more minutes of transcript below
Thousands of transcripts fetched by people building searchable podcast archives
Fetch the whole transcript
The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000792376050