Anthropic releases Opus 4.8 with new ‘dynamic workflow’ tool
Anthropic released the latest version of their frontier model, Opus 4.8, just 41 days after the release of the previous version. That release showed a huge jump in performance and when I wrote about it I suspected that it was actually more like a preview of Opus 5.
Opus 4.8 is more of an incremental improvement on Opus 4.7 but the killer feature is how it manages mistakes - it’s far more likely to flag uncertainty about its work and less likely to make unsupported claims. Both of these things have been a big issue for frontier AI models, and are particularly problematic for models that power agentic workflows where you need self-reflection and honesty.
As part of the release, Anthropic also commented on Mythos, their incredibly powerful cyber-security model and it sounds like it’s nearing general release. They’ve been making good progress on the required safeguards and expect to be able to release it to all their customers in the coming weeks. If that’s the case I suspect we will be seeing Claude Opus/Sonnet 5 towards the end of the summer.
Anthropic reaches valuation of $965bn, beating OpenAI to become world’s most valuable AI firm
The good thing about funding rounds is that it puts an agreed valuation on a private company. In Anthropic’s recent round of funding they raised $65bn at a valuation of $965bn. This is the 8th funding round Anthropic have completed, with the first one on May 2021 raising ‘only’ $124m at a valuation of $845m, meaning they have over 10x their valuation in just 5 years.
Anthropic have been on an unbelievable growth tear over the last 6 months. Whilst Anthropic’s revenue is rarely publicly declared The Information has reported that Anthropic are now generating an annual revenue run rate of $45bn, up from roughly $1bn 18 months ago. In Q1 2026 Anthropic predicted their revenue would grow 10x, but it in fact grew 80x - madness!
As Simon Willison has written, Anthropic and OpenAI have found product-market fit which is what’s driving these revenues. That fit is coding agents with enterprise customers paying huge amounts for API use. The medium term bet is that they will be able to extend this success beyond coding agents to general purpose knowledge work agents and that’s when usage (and revenue) will really explode.
If Anthropic and OpenAI are able to crack that then 80x revenue growth in a single quarter will look like pocket change. I think we’re in for a bit of a wild ride over the next couple of years!
The Guardian | The Information | Fortune | Simon Willison
Are consumers ready for humanoid robots?
Great consumer report from the FT on the current state of humanoid robotics and where it’s headed. Interesting stats they shared:
Investment has tripled over the last 2 years to over $40bn
Shipments are expected to grow from 90k to 1.2m in 2030
The market will be worth $200bn by 2035
There’s still a long way to go but a lot of investors and start ups are betting on being able to prove that an affordable general-purpose humanoid robot will be the next big consumer tech product.
School of Football
Great promo for Hyundai (who own Boston Dynamics)’s sponsorship of this year’s World Cup. I not for a second suggesting that a robot can learn how to play football just by watching some footage, but it will be fun to see how they get on teaching Atlas to play!
Web 4.0
AI Ethics News
Long Reads
One Useful Thing - Choosing to Stay Human
Benedict Evans - Predicting AI job exposure
Stratechery - The SpaceX IPO and Data Centers in Space
“The future is already here, it’s just not evenly distributed.“
William Gibson