OpenAI Is So Back... GPT 5.6 Soul First Look

[00:01] confirmed it probably wouldn't kill us all, Open AAI was granted permission to unleash the GBT 5.6 family of models onto the public. It comes in three sizes models named in New Jersey style Italian, Luna Terra, and the flagship

[00:17] Gigabrain Soul. A model that shatters the latest Trust Me Bro benchmarks, and quite possibly outsmarts Claude Fable, Mythos, and every other model that came before it. If you're an agentic engineer, formerly known as a vibe

[00:29] coder, this is a huge deal because soul tops the agentic coding leaderboard. It has a new ultra mode that can spawn an army of sub aents. And in my experience, to getting stuff done. But the timing could not be more sus because just last

[00:42] week, Anthropic's Fable 5 was brought back to life after Jailbreakers turned simultaneously Elon just released Grock 4.5, which might be done by comparison, but uses a tiny fraction of the tokens as Soul or Fable. In today's video,

[00:57] we'll get back on the AI hype train and find out if these new models are true game changers. It is July 10th, 2026, and you're watching the code report. To understand the launch of GPT 5.6, you need to first understand that Frontier

[01:11] AI models must now go to the DMV before you can drive them. Back on June 2nd, the president signed an executive order asking AI labs like OpenAI and Anthropic to voluntarily hand over their most powerful models to the government for up

[01:25] to 30 days to review before release. This process, of course, is totally volunteer, you might as well throw your company in the wood chipper. So, when GPT 5.6 dropped a few weeks ago on June 26th, it went to just roughly 20 trusted

[01:39] partners whose participation has been shared with the government. And now that you feel safe enough, let's go ahead and take a look at GPT 5.6 Soul. Basically, OpenAI's new strategy isn't about making the base model smarter. It's about

[01:51] giving the model an entire virtual sweat shop of sub agents that can be whipped like slaves into building stupid apps that nobody will ever use, like Horse attract investors. But with this new family of models, it comes two knobs

[02:04] Reasoning, which is basically like Claude's deep thinking mode, and ultra mode, which tells the model to spawn a team of AI slaves to tackle problems in parallel. That's especially useful for programmers, if we can even still call

[02:17] agent writing some React components while another handles a database while the third one fails to build a beautiful UI with CSS. And it makes all the multi- aent orchestration startups out there obsolete, except of course the ones that

[02:30] a look at Terminal Bench 2.1, which is a it tests real command line workflows versus pointless middle school math, we can see that GPT 5.6 Soul beats Claude Mythos 5. But on Soul Ultra mode, it

[02:44] chubs up to a ridiculous 91.9%. cyber security, it still just barely underperforms Claude Mythos on the exploit gem benchmark. One thing that's interesting though is that they left out

[02:57] swebbench pro which is a benchmark on real GitHub issues with real code bases. benchmark and OpenAI simply didn't publish a score which in my opinion underperforming there. In addition, Meter, a nonprofit AI evaluator,

[03:13] detected an unusually high rate of cheating in their initial evaluation. It or shortcut metrics to avoid doing actual work. As some people might call that cheating, but others might say it's working smarter and not harder. The only

[03:26] way to truly analyze it is to analyze it yourself. But the big question for everybody right now is which one is better, Claude Fable or GPT 5.6 Soul. I Neither of them have ever sponsored this channel. And what I can tell you is that

[03:40] Arguing about which one is more intelligent is like arguing about who's Ronaldo Messi or Erling Holland. They're both so much better than us at soccer in much better than us at anything involving intelligence. But Soul is

[03:55] about half the price and tends to get the job done faster. It's like a contractor who shows up with six guys and finishes the job in record time. A really good contractor who goes very slow but gets the job done right and

[04:08] Ultimately, it just comes down to using the right tool for the job, which is why you also need to check out Blacksmith, the sponsor of today's video. It's a drop-in replacement for GitHub runners that lets you run your GitHub actions

[04:21] twice as fast while costing 75% less. That might sound too good to be true, speed by running your actions on bare metal gaming CPUs with the highest single core performance. But Blacksmith also makes your GitHub actions fully

[04:35] observable, so you can quickly track down what's happening in your CI which is critical when you have agents writing thousands of lines of code with Blacksmith for free today at the link

[04:48] per month when you sign up. This has been the Code Report. Thanks for watching and I will see you in the next one.