Pulse Daily — Thursday, October 8, 2026

Amber, you're watching the model releases this morning.

What's the headline?

Thanks, Ido.

The headline from the model front is efficiency, and it has a price tag attached.

Anthropic today launched Claude Haiku 5.5, pricing it at $0.10 per million input tokens and $0.50 per million output tokens, which works out to roughly 75% cheaper than the previous Haiku 4.5. That is a dramatic cost reduction, and critically, Haiku 5.5 now matches OpenAI's GPT-6 Luna on price.

But the real news isn't just the cost.

It's the capability attached to that cost.

Haiku 5.5 becomes the first Haiku model with effort controls, which lets developers dial compute intensity up or down depending on the task.

It scored 72.4% on OS World 2.1 for computer use.

For context, that's an agentic benchmark.

It measures how well a model operates a real desktop environment, So you're getting frontier-class computer use capability at a price point that high-volume applications like summarization and classification can actually afford.

This is the race to the bottom that, paradoxically, is expanding what's possible at the top.

The cheapest models are now doing things the expensive ones couldn't do a year ago.

And while Anthropic was cutting prices, OpenAI was cutting access barriers.

The company today rolled GPT-6 to all ChatGPT subscription tiers – Pro, Plus, Business, and Enterprise – via the Chat tab, with an interface update being called the Intelligent UI.

That is a significant democratization moment.

GPT-6 had been available to a limited audience, and now it's the default experience for millions of paying users.

To put the revenue picture in context, OpenAI's annual recurring revenue has now approached $70 billion, up more than 70% since the start of the third quarter of this year.

That is an almost incomprehensible growth curve.

The company is simultaneously targeting a $30 billion bridge round at a $1.4 trillion valuation and has pushed its IPO timeline to 2027.

The capital structure around this company is becoming as complex as the models it's building.

And speaking of capital structure, the chip financing story today is extraordinary.

Broadcom has been working to arrange more than $50 billion in financing for OpenAI's custom AI chip, with Oracle simultaneously in talks on financing for a separate major chip purchase, and Apollo, Blackstone, and Goldman Sachs all named as lenders in these discussions.

The numbers are so large they require a moment to absorb.

This is not software infrastructure spending.