striking

The fixed consumption plans are offering several times their worth compared to API pricing with completely free cache reads: https://she-llac.com/claude-limits

I look at that and think that they must be losing money hand over fist on something like this, not that this shows what their margins are like. If their margins are like this then I don't see why they'd be raising money and shuffling it around in circles.

If it's really that easy to get better coding performance, why haven't the chinese labs replicated it?

Nobody said it would be easy! I just think it's possible, and that presumably they will get around to doing it at some point.

I'm arguing we can't trust retail prices because the marginal pricing isn't meaningfully connected to it anyway.

But if we have to look at what we think margins might look like, DeepSeek continues to host v4 Flash at the existing price despite competitors beating it in price (https://openrouter.ai/deepseek/deepseek-v4-flash), so there's at least one example of a Chinese lab charging a predetermined price despite competition. And no one but Moonshot is hosting Kimi K3 yet (https://openrouter.ai/moonshotai/kimi-k3). Perhaps there's room in the market for those who release their models to make margin on them.

And I believe my Composer example speaks for itself. The open models are behind but there's tangible proof they can be tuned for pareto frontier efficiency. See "Cost per Task" at https://artificialanalysis.ai/agents/coding-agents.

Sure, let's have a look...

I highly doubt that Chinese models are cheaper to serve on a marginal cost basis, they just seem cheaper because Anthropic and OpenAI are so supply constrained that they are charging far more than they would if there were sufficient supply to meet the demand for intelligence. [emphasis mine]

I guess I'm missing the part of this article where they bring hard numbers in to back up the argument here. What work was attempted? https://cursor.com/evals shows the previous generation of open models (Kimi K2.7) trading blows with the others, cost effectively. Composer 2.5 is itself a fine-tune of K2.7, and it's apparently quite token efficient, so why would it be impossible for a Chinese lab to achieve something similar? GLM 5.2 Max is also ranked above the lower end OpenAI models and is not far off in price.

It's weird to have this entire discussion about tokenomics without mention of the circular financing and debt raised by labs in the West, which can then essentially give away their capacity to end users. OpenAI giving away quota resets to subscribers like candy on Halloween while their compute partner Oracle's bonds is reevaluated to be one grade above junk? How?

I don't think you can make an argument about the future one way or another by arguing using the listed prices. The math is not internally consistent enough for it.

https://archive.is/KPans

I've played through the whole of story mode with no more than a few rare world streaming issues and that was a couple years ago. You could give it another shot, I recommend binding the gyro to one of the back grip buttons for precise aim.

Better yet, get the original release of GTA: San Andreas. It holds up to several replays in a way that GTA V does not imo.

Movi M10 uppper brace and bolt holder

They're cat ears. You can just admit it. We won't judge.

It just is what it is. Sometimes you want to write the obvious query without the DB getting in your way, and other times you want to know as soon as possible that you're doing something that won't scale under exponential load. At this point in my career I prefer the latter, but the former will always have a special place in my heart.

Maybe one day I’ll learn to read a query plan.

With SQLite's `.expert` mode you can delay that day a little longer: https://www.sqlite.org/cli.html#index_recommendations_sqlite...

  sqlite> CREATE TABLE x1(a, b, c);                  -- Create table in database 
  sqlite> .expert
  sqlite> SELECT * FROM x1 WHERE a=? AND b>?;        -- Analyze this SELECT 
  CREATE INDEX x1_idx_000123a7 ON x1(a, b);

  0|0|0|SEARCH TABLE x1 USING INDEX x1_idx_000123a7 (a=? AND b>?)

  sqlite> CREATE INDEX x1ab ON x1(a, b);             -- Create the recommended index 
  sqlite> .expert
  sqlite> SELECT * FROM x1 WHERE a=? AND b>?;        -- Re-analyze the same SELECT 
  (no new indexes)

  0|0|0|SEARCH TABLE x1 USING INDEX x1ab (a=? AND b>?)

My approach so far has been to just do these cleanup operations in small batches so that I don’t need to do database queries that take more than 5 seconds to run. This whole experience has given me more of an appreciation for why someone might want to use a “real” database like Postgres which can have more than one writer at the same time though.

The advice for those " “real” " databases is generally to also do cleanup operations in small batches, they just tend to make it less obvious you're doing something unperformant in the smaller case. You're more right than you thought!

https://www.bloomberg.com/news/articles/2026-07-16/google-ge...

I believe it's 5 minutes on API pricing by default, though you can turn off caching or force it to 1 hour. Subscriptions are special, seemingly because Anthropic doesn't want to expose casual users to all of the tokenomics.

Anthropic's cache expires after 1 hour when using subscription endpoints, and for those cached tokens cache reads are free. It's generous (compared to API pricing) but it's not 100% free.

I don't see any forward looking timelines on the page, is this the right link?

https://dictionary.cambridge.org/us/dictionary/english/disti...

"Fulfil" is the same way

This might be good advice for people who haven't already had the rest of the living optimized out of their lives. But not everyone is so lucky...

Also, the article kind of agrees with what you're saying, even if you don't realize it. Any exercise, even if it's not organized like a weight lifting regimen might be, might be enough to keep you healthy. I think leaning into that could make exercise more genuinely enjoyable for many more people, instead of just an exercise in optimization.

I got a low profile ortho 40% called Technik from Boardsource: https://www.deskhero.ca/products/technik-by-boardsource-xyz

Unfortunately seems to have entirely disappeared from their website despite being a really lovely product that I've daily driven for years. If anyone finds a replacement I could recommend to friends I'd love to know about it.

I think there is in fact a discrepancy between displaying flags and advocating directly for gendered violence:

Debian contributor "NoisyCoil" said that they had wanted to argue in favor of keeping the packages, but after looking at the content they had decided against it:

> I went peeking at the package and, unless I'm completely missing something, the second offensive Italian fortune says that women's "no"s should be interpreted as "yes", while the third one explicitly calls for violence on women [1]. Like, it literally says women should be beaten on a regular basis. I'm afraid I can't help you here, sorry.

from https://lwn.net/Articles/1031750/, linked in the fine article

https://en.wikipedia.org/wiki/American_exceptionalism

Once you see it as a throughline of your conversations with people who have been steeped in the American culture for so long, it is hard to unsee.

https://xcancel.com/AnthropicAI/status/2070665903440871779

Anthropic @AnthropicAI Jun 27, 2026 · 12:29 AM UTC

Since June 12, we’ve been working closely with the US government to restore access to Claude Mythos 5 and Fable 5. Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure.

We’re restoring access for these organizations quickly, and we’re continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.

NYT seems to have finally figured out how to block archive-today.

I mean. I looked up https://en.wikipedia.org/wiki/Plombir

The standard at which it was made appears to be a historical fact.

Yep. The only place I have had anything like this is at the University of Connecticut dairy farm. They make the ice cream on campus from dairy cows that live on campus. And it's lovingly made by bushy-tailed, bright-eyed students, so the technique is immaculate as well.

There are some facsimiles where I live now but nothing could ever replace the real thing.

I've not personally had it but from my experience with fresh high quality high milk fat dairy (which I believe Soviet ice cream to be), the richness, high fat content, lack of air bubbles and stabilizers help keep it from melting.

And when we do let them self-regulate (like Boeing) shit hits the fan a few years later (737 MAX et al.) like clockwork.

Unfortunately I have just as little trust in this instance of the US government as I do the corporations. Hopefully it's only two more years of this.

https://en.wikipedia.org/wiki/DRAM_price_fixing_scandal

To date, five manufacturers have pleaded guilty to their involvement in an international price-fixing conspiracy between July 1, 1998, and June 15, 2002, including Hynix, Infineon, Micron Technology, Samsung, and Elpida.

It is history; we have not learned; we are doomed to repeat it.

From https://www.newsweek.com/brian-thompson-muder-health-insuran...

The fatal shooting of UnitedHealthcare CEO Brian Thompson has prompted healthcare executives to say they will address growing frustrations among Americans struggling with access to and costs of medical care.

From https://www.cnn.com/2025/06/23/health/health-insurers-preapp...

Months after the killing of a top health insurance executive unleashed Americans’ pent-up anger over denials of medical care, the industry announced Monday that it will take action to “streamline, simplify and reduce” the preapproval process.

However, from https://www.cnn.com/2025/12/04/health/insurers-prior-authori...

However, multiple provider associations and patient advocacy groups interviewed by CNN say that little, if anything, has changed over the past year.

So, hard to say for sure.

I mean it kind of already is in harnesses. Codex and Claude Code both have subagent tools. You could probably get a similar token output cut just by asking Claude Code to run all commands with Haiku as a summarizing subagent.

Take for example https://github.com/rust-stdx/stdx/tree/main/itoa. Its licenses and copyright information have been stripped. You are permitted to make copies of code under the MIT license, but the license also includes:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

If the original repo were to disappear, it would be important to know who wrote the code and what the license actually is.

I'd be pretty pissed off if my LLM told me the only solution it'd be willing to implement to fix my code is to rewrite it in Rust. No way I'd pay for a model that refuses to fix bugs in the language given, especially because maybe I might not have the ability to convince other stakeholders to change it.

1. Spoilage claims are not usually covered by insurance if the power interruption starts at the utility company instead of in your building. It's pretty rare that anyone knows to ask for that particular endorsement, and you're SoL without it.

2. You don't think everyone else was thinking the same thing?

Also, you can just have sympathy for people even if it were their own mistake. It doesn't cost you anything.

You can also use one of the accessibility shortcuts to trigger it quickly (both volume buttons held, triple tap, two fingers from bottom) if so configured