Yeah at the point that Opus has read enough context to hand off a detailed plan, implementing the plan may not be that many more tokens.

Also having spent the weekend with the supposedly great Deepseek Flash 4.1 1017, I wonder if the people singing its praises have used recent frontier models. DS Flash feels like maybe Sonnet 4.5-4.6. Absolutely falls apart at things Opus/Sol currently excel at with low reasoning, and even has basic problems with failing tool calls, which hasnt been an issue with Anthropic/OpenAI in forever.

Could be a configuration issue with the provider I tried but honestly, I didnt get much better out of GLM5.3-flash. GLM5.3 and Kimi3 are much better but since most inference providers cant take a loss they are more expensive than just using Opus/Sol on subs, which are almost certainly currently a loss leader for their respective API offeringss.

> Huge, generational mistake for the USA to be leaning so hard into fossil fuels right now and actively trying to prevent buildout of solar and wind power.

If the US is trying that, they are failing pretty hard - 88% of new power in the US in 2025 was renewables, and 2026 is estimated to be above 90%

I'd rank Sol 6.1 only slightly behind Opus 5.5 in intelligence and both significantly ahead of anything else I've used (though cheaper/faster models are totally fine for a lot of tasks). My only real annoyance is 6.1 Sol is one of the slowest models I've used, which does make Opus 5.5 more pleasant to use.

> and both significantly ahead of anything else I've used

Really? I just don't see it, working with Sol has it constantly do stupid, dangerous things I never asked for, e.g. "I’m sorry. You asked whether SSH was possible, and I changed the firewall without being asked". I asked literal question, it proceeds to answer and changes iptable settings on a device I am working on, risking a complete lockout from it.

It is genuinely dangerous and has to be looked at all the time. Fable, Astra and Opus 5.5, but also Deepseek 4.1 are ahead of it. I had DS working on the same project as Sol and DS was correcting Sol's work routinely.

There is a one click update in the WebUI now, no need to flash random bins

This is good news then. I divested from the OpenWRT ecosystem about 7 years ago because of an experience similar to my example.

Case in point was an Asus RT-56N router that I had to replace with ISP wifi.

GLM 5.3 is roughly Opus class and is much, much cheaper

It’s more expensive than 6.1-sol though? At least according to benchmarks.

I think at this point comparing raw token pricing is often apples to oranges.

Those "price per task" benchmarks seem to pre-suppose the existence well-defined requirements and acceptance criteria.

When that isn't available, and the task involves "understanding the problem", "deciding what I'm actually after" etc. iterations, I find price-per-token to still be a better predictor of cost.

Opus 5 class yes, but opus 5.5 is functionally mythos tier

Opus 5.5 class?

No, since that's a moving target. But the large GLM models are likely in a similar model weight class to the Opus family, just a few months behind

No, and anyone who says otherwise is being silly.

It’s still damned capable though, I’ve done a crazy amount of good work with 5.3 Flash.

> I think we have the moral imperative to document all "out of distribution" information somewhere

This is more or less why writing was invented thousands of years ago, the amount of knowledge exceeds human memory

That’s not the point I made actually. Most writing is redundant. Our moral imperative should only be to document the out of distribution information - stuff that you can’t elicit from model.

If your writing is compressible, you are doing it wrong.

EDIT: I see your point. Same mechanism, different substrate.

Have used decision models the same way: Had LLM generate/tune questions for Clef to label/extract emails with {useful work thing} content

Too early to say how useful this is versus existing techniques, the cool thing is this is the first thing I've found local models genuinely useful for. I think I was able to process 50,000 email questions per hour or so with Clef flash on Strix Halo with a custom engine. A modern gaming GPU would likely be much faster

> which is weird when Anthropic have supposedly claimed two profitable quarters in 2026

Anthropic claimed to be profitable if you dont count some of their largest expenses (like model training). Not the sort of profitable where revenue actually exceeds total expenses

You sure they're not including training costs in those probability claims? I've not seen confirmation one way or the other on that.

Remember, when you see “EBITA”’think “bullshit earnings”

Not a new revelation, but I think its funny that there is a inflate_signal_strength_bool setting and many carriers set it to true

It almost reminds me of the 2010s era US market, when both AT&T and T-Mobile passed off their 3G coverage as "4G" even though it wasn't. If you had a suitable phone and got onto a T-Mobile tower that supported DC-HSPA (dual carrier) it was 4G-ish, but it certainly wasn't LTE.

I'm just glad we don't have the GSM vs. CDMA games from LTE onwards. Of course, that doesn't guarantee your phone supports a given carrier's frequencies.

What was quite bizarre with the 4G deceit was that at least in SF, AT&T were pretty aggressive with the real LTE rollout so they had to rebrand that to 4G LTE.

Compute and payroll