40% cheaper: https://platform.claude.com/docs/en/about-claude/pricing

But what about real price in real agentic use? For example, Opus 4.5 was more expensive per token than Sonnet 4.5, but it used a lot less tokens so final price per completed task was very close between the two, with Opus sometimes ending up cheaper

How does it work exactly? How this model is cheaper and has the same perf as Opus 4.5?

Distilling from a teacher (Opus 4.5) and scaling RL more.

So less parameters but "better" weights?

this is called progress

I'm asking technically how progress works. What is actually being improved here

mostly cost of hardware going down. as models scale, nvidia produces a new hardware generation that outputs more tokens per watt, but those speed gains get eaten by the fact that the model is bigger ie. more expensive to serve.

Also we have no clue whether Anthropics inference margin is compressing or not and they just want to maintain the price.

Or, we can bleed out cash for a very long time.

How can you determine whether it's as good as Opus 4.5 within minutes of release? The quantitative metrics don't seem to mean much anymore. Noticing qualitative differences seems like it would take dozens of conversations and perhaps days to weeks of use before you can reliably determine the model's quality.

Just look at the testimonials at the bottom of introduction page, there are at least a dozen companies such as Replit, Cursor, and Github that have early access. Perhaps the GP is an employee of one of these companies.

Given that the price remains the same as Sonnet 4.5, this is the first time I've been tempted to lower my default model choice.

If it maintains the same price (with Anthropic tends to do or undercuts themselves) then this would be 1/3rd of the price of Opus.

Edit: Yep, same price. "Pricing remains the same as Sonnet 4.5, starting at $3/$15 per million tokens."

3 is not 1/3 of 5 tho. Opus costs $5/$25

> That's a long document.

Probably written by LLMs, for LLMs