Andy Halliday

And it's an open source model, meaning that, you know, as Beth pointed out yesterday, it doesn't mean you can run it on your local machine.

But the scientists who are working on new models and wanting to make distillations of models can now download this on a sort of a lab level infrastructure and they can do distillations of that and so on.

So what you cannot do with Gemini or OpenAI or XAI or any of those closed models, you can't do that.

Well, they made a big splash with GLM5.

Then yesterday, Minimax launched its M2.5 model.

Yes.

And it has performance that rivals Opus 4.6, Claude Opus 4.6 in coding and agentic tasks.

The model uses a 10 billion activated parameter model of sparsity activation and supports approximately 100 tokens per second.

It's designed for full stack development access across all the different platforms.

And shares in that company, Minimax, surged 24% in Hong Kong trading when they released this thing with its

benchmark performance uh it's an amazing model so the chinese open source models glm5 from zed and now minimax 2.5 they're really right up there alongside the other major models right the headline for minimax is like it's a buck an hour pricing

Which undermines the revenue models of the other major model companies.

But the one other big news is that Anthropic completed their funding round.

They raised, I think, $30 billion.

$30 billion.

Their valuation in this round was $380 billion.

But behind that announcement about the completion of this new funding round for Anthropic is that their current run rate as of today is $14 billion annually.

which is 10 times the revenue from just a year ago, something like that.

Their ARR for Claude Code alone is at $2.5 billion.

That's annual recurring revenue, $2.5 billion being spent on Claude Code.