Vision Pro would disagree

They have the resources to be able to do what Jobs described as “hobby projects” (he was referring to the Apple TV which definitely is the best streaming interface for video available, albeit one which is on the expensive side). They did kill their car project after spending a decade on it (and reportedly having thousands of people working on it).

Now they have them, in the 1990's they were almost bankrupt with similar projects, and got lucky with NeXT "buying" them.

Apple-NeXT is probably the most successful acquhire in the history of modern business.

6 Sol is the new Terra with 50 percent discount vs 5.6 Sol.

Also 6 Astra Mini would be out soon which would be 5.6 Sol pricing?

I am curious -- what is the basis for claiming that Astra 6 Mini will be out? It doesn't make sense to me. It would only further antagonize and confuse users who seek some predictability from a model family.

Yup more like a 5.7 than a 6

Number is generation Name is the size (Luna smallest to Astra largest)

Then what is Astra high-extra high-Ultra? That’s effort within each tier?

Yes that is number of reasoning tokens used.

Performance increases both with larger model (Luna vs Sol)

And with more reasoning (low vs xhigh)

Exactly

Sol 6 is also dumber than Sol 5.6 on some tasks (DeepSWE)

A vs B

Should be B vs A correct?

Else it's confusing

6 Sol Performs worse than 5.6 Sol at DeepSwe?

Wierd!!

Opus 5.5 seems better? Can someone attach both scores

Not the direct competitor to Opus 5.5, cuz 6 Sol is 50% cheaper.

Same price on Cache Reads 0.2/M So won't be 50 percent cheaper, more like 25% cheaper assuming half cost is cache read.

that might depend on usecase, half of my cost is cache reads usally

DeepSeek v4.1 Flash 31.2%

Source: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash#compa...

I have been using it today the whole day and it is definitely better than Luna. Great that bench agrees.

Can you tell something about your tasks? I am pondering both models for cost saving.

Yeah DeepSeek V4.1 beats this by 15 percent (4 points) on terminal bench 4.0