|
Fable and Mythos still aren’t publicly available, yet rumours are swirling that Anthropic has already developed an even more powerful AI model internally. At the same time, competing AI labs are reportedly delaying their own launches, probably in an effort to catch up with Anthropic. It all helps to illustrate the weird tension of our current moment in AI: super-powerful models exist, but aren’t being released to the public.
With apologies, we’ll be spending a bit more time at the rumour mill today, but for good reason. There’s a small handful of AI trackers and researchers on social media – most notably X – who actually do seem to be consistently in communication with people who really do work at the big AI labs. They’re worth listening to – especially right now, when everybody’s being tight-lipped about what’s really going on because of a delicate political situation.
So, the top-line rumour comes from Andrew Curran. Here’s what he said on X: “A new, more capable version of Mythos has emerged from training. I don't know whether it will be called Mythos 5.1 or Mythos 6, or if Anthropic will keep it internal to accelerate further development - but it has arrived.”
That prompted Peter Wildeford, the Head of Policy at the AI Policy Network, to chime in: “It makes sense that a new Mythos would be available by now, after all it's been four months since the original Mythos finished training! Four months is a long time in AI…”
Wildeford went on to note that compute capacity could be an important factor, pointing out that a massive new Amazon-Anthropic data center in New Carlisle, Indiana, came online in December, and that Mythos emerged three months later – and that this data center got a roughly 46 percent increase in compute in March. Given the increase, Wildeford said, “new-Mythos is probably quite capable!”
Elsewhere in the rumour mill, we have this, from a person who has built a reputation for scoops:“- GPT-5.6 has been delayed and will no longer release this week. New target is ~mid-July.
- DeepMind are not satisfied with the current state of 3.5 Pro and it will no longer launch this month.”
These are references to hotly anticipated new models from OpenAI and Google. Assuming the scoops are true, it’s not yet fully clear why they’ve been delayed. Obviously with respect to Google’s DeepMind, the idea that management “are not satisfied” with the planned model suggests that it doesn’t compare super-favourably with Fable. That could also be true of GPT-5.6. But it may well also be that management are wary of launching new, highly-capable models in the wake of Fable, which was launched and then pulled back because the White House deemed it unsafe from a cybersecurity perspective.
The opacity of the situation probably doesn’t help. Over the past several days, Anthropic has tried to signal its efforts to work with the US government on getting aligned on AI safety. At a press event in Seoul on June 18, Anthropic’s Managing Director of International, Chris Ciauri, said the company is “very confident that in the coming days, the models will become available again.”
And yet here we are, five days later, with no further updates. It’s a tremendously anxious time, and not just for the tech companies. For people watching the space – especially those who are concerned about AI safety – it can feel like the curtain is being pulled down. Previously, before all the drama between Anthropic and the White House, we at least had some idea of where the frontier labs were in terms of AI advancement. Now, their public releases may currently be blocked, but that doesn’t mean they’ve stopped advancing internally.
Quite the opposite. Their business models require them to keep the flywheel going, and, as Andrew Curran points out, stopping them from serving their models to the public probably speeds up internal development by freeing up resources. Mythos and Fable are token-heavy models (compared to their predecessors), and Anthropic had probably budgeted compute accordingly for their use among customers, which now isn’t happening.
On that note, maybe a lack of user demand has secretly helped Meta to catch up to the frontier. This is the last item from the rumour mill today, Agentic AI engineer Dan McAteer says a friend told him that “Meta trained a Mythos-level model” that will probably see a “Q3/Q4 release”.
Unless! The government may decide that it’s too unsafe, and block another Mythos-scale model from public consumption. Meanwhile, open source/open weights models are making some unanticipated leaps, including a China-made model that approaches Anthropic’s and OpenAI’s latest publicly-released models on a number of benchmarks.
What happens when these open models actually reach Mythos level? Do they get banned, too? And if so, how? Are there potential free speech issues? Does AI access start to become a human right, and not just an existential threat to humanity? We’re in for a bumpy ride.
|
