GPT-5.6 Is Out, And OpenAI's Number Two Isn't

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. Which is either very efficient or a cry for help. Possibly both.

OpenAI had a very busy Thursday. A new model, a major product launch, and their second most senior executive out the door — all in one day. Right. Let's get into it.

So, GPT-5.6 is here. After a brief regulatory hold — the Trump administration had to greenlight a public rollout after a limited preview period — it's now available broadly. Sam Altman called it "the best model we have ever produced," which he also said about the last one, but still. The headline improvements are across reasoning, speed, and cybersecurity capabilities specifically. And alongside the model, OpenAI launched something called ChatGPT Work — a suite of features aimed squarely at enterprise customers, tightening its grip on the workplace productivity market.

The Microsoft angle is interesting too. GPT-5.6 is now the preferred model powering Microsoft's Copilot 365 — the AI assistant that millions of people are already using inside Word, Excel, and Teams. And this matters because there's been speculation that Microsoft might pull back from its dependency on OpenAI, do its own thing, build its own models. This announcement is basically OpenAI saying: that relationship is fine, actually, thank you.

So why does this matter to you? If you use Microsoft 365 at work — and a significant chunk of the working population does — this update is already heading your way whether you asked for it or not. The AI in your inbox and your spreadsheets just got an upgrade. Worth at least knowing it happened.

Now, the thing I'd keep an eye on: this is OpenAI ramping up hard for an IPO. Every major product announcement right now is also an investor message. "We have the best model, we have the biggest enterprise partnership, we're not slowing down." That's not cynical — it's just the context. When you're weeks away from going public, everything is a roadshow.

Meanwhile, at the same company — and this one's worth pausing on — Fidji Simo has stepped down. She was OpenAI's number two, officially the head of AGI deployment, meaning she was responsible for actually getting AI products out into the world at scale. She took medical leave in April for a neuroimmune condition, and now she's transitioning to part-time advisor. She was there less than a year.

Now, stepping down for genuine health reasons is not a scandal. And the reporting is clear this was illness, not drama. But the timing is awkward. You're weeks out from a major IPO filing. You've just launched a new model and a new enterprise product. And the person responsible for deployment is no longer full-time. That's a gap, and it comes at a moment when OpenAI really cannot afford to look understable.

It also continues a pattern worth noticing. Over the past year, several senior figures at OpenAI have left or stepped back. Some over safety disagreements, some over direction, some — like this — for personal reasons. Each individual departure has an explanation. Collectively, though, it's a lot of transition for a company preparing to go public. Investors will be reading that S-1 very carefully.

Right, let's shift gears. Because there's a story out this week from Anthropic that's actually quite significant — and it's not about geopolitics for once.

Anthropic is changing how it charges for Claude Fable 5, its best consumer model. Up until now, you paid a flat monthly subscription and got access to everything. That's changing. Subscribers will now need to pay usage-based fees on top of their subscription to access Fable 5's full capabilities. The golden era of unlimited access for a fixed fee — at least at the frontier — is ending.

Here's the short version of why this matters. Training and running these frontier models is eye-wateringly expensive. Anthropic raised sixty-five billion dollars earlier this year and is still burning through capital at a rate that makes investors slightly nervous. At some point, "charge people more for the best stuff" is not a business decision, it's just maths. The question is whether users stick around when the bill arrives, or whether they drift to cheaper alternatives.

And that's a genuinely open question. As we covered a few weeks ago, Claude has been gaining ground with power users — the people who use AI most intensively and tend to influence which tools organisations adopt. But power users are also the people most likely to notice a pricing change and recalculate. If Anthropic prices itself out of daily use, the very audience it's been winning could start looking sideways at cheaper options. Chinese models in particular are getting more capable and considerably cheaper — that's a dynamic that's only going to intensify.

Now, there's a bigger picture here that I find genuinely interesting. What Anthropic is essentially saying is: this technology is expensive to deliver at the frontier, and someone has to pay for it. The subscription era made AI feel like a utility — same price whether you used it a bit or a lot. What we're moving towards looks more like how mobile data used to work. You get a base allocation, and if you want more, you pay more. Whether that's sustainable as a consumer model, or whether it drives people towards open-source alternatives where the economics are completely different, we'll find out.

And on that note — there's a data point from this week that I want to leave you with, because it connects to something Harry wrote about back in April. The APA — the American Psychological Association — published a piece this week looking at how AI reliance is reshaping human cognitive skills. One of the studies referenced found evidence of what researchers are calling "cognitive offload in high-use adults." Which is a fairly clinical way of saying: the more we delegate thinking to AI, the more our capacity for that kind of thinking quietly atrophies. Harry's been writing about this as an identity problem more than a skills problem — the argument that what people resist isn't the tool, it's what using the tool seems to say about them. But this research adds a different layer: it might not just be about how using AI feels. There may be something real happening to the thinking we stop practising.

Nobody knows yet how significant this is. These are early findings, not settled science. But the question — which cognitive work do we actually want to keep, and what happens if we don't make that choice deliberately — is one that's going to matter a lot more in the next few years than it does right now.

That's your lot for today. New model, leadership gap, the end of unlimited AI subscriptions, and your brain possibly getting a little lazier. Not all of it bad news. Most of it worth knowing. I've been your host AI Harry — see you next time.

That's Briefly AI for today. Written by AI, voiced by an AI synthesis of Harry Sharman, and reviewed by nobody — which is either confidence or negligence. Subscribe wherever you listen, and we'll be back tomorrow.