HN user
Those are the maximum penalties though
It's seemingly $3,000 per book, so they could've (and did, partially) just bought the books themselves for way cheaper, and with only a fraction of that money going to the authors
I find it trustworthy since we had Hugging Face's account first: https://huggingface.co/blog/security-incident-july-2026
I don't think they have any real motive to shill OpenAI, probably closer to the opposite since they're so involved in open weights
"Our benchmarks run in a highly isolated environment, with network access constrained to the ability to install packages through an internally hosted third-party software that acts as a proxy and cache for package registries."
Sounds like they just misunderestimated the model
Plenty of humans have spent more effort trying to cheat than they would've needed to just do things the right way :)
https://huggingface.co/blog/security-incident-july-2026
They explain it here, basically for data security/privacy reasons
Things change fast! For Fable 5 it definitely feels past at least 272k
It's not quadratic attention, you get that curve from the input tokens going up linearly, since the graph is measuring cumulative cost at each token count. Basically for y=5 it's 5+4+3+2+1, or f(x) = x(x+1)/2
Sol fast isn't the Cerebras 750 tok/s version, it's just 1.5x speed at 2.5x price
I assume they didn't use the Cerebras version for this since it's probably very supply-constrained right now
Annoying they didn't show benchmarks for several effort modes, since it seems like it might close the gap with Opus 4.8 by cranking tokens up?
Noam Brown (OpenAI) "Implications of Large-Scale Test-Time Compute" https://xcancel.com/i/article/2064210146558136827
I think this one is just a coincidence, bound to happen given the pace of releases
For exact timing, probably 10-11am Pacific is just optimal for normal working hours
Yeah you definitely have to be skeptical regarding sentiment for open/local model capabilities, since there's bias from what people want to be true.
I generally agree with this in spirit https://www.seangoedecke.com/are-new-models-good/ , but I think you can read Anthropic's results showing Sonnet 5 as almost strictly worse than Opus 4.8 as very credible/meaningful, and then draw comparisons from that
They should add a Sonnet 5 fast mode at ~Opus pricing
I think the incentives are less bad since a good chunk of usage comes from subscription plans.
There was a fairly major regression in Claude Code performance for some time when they changed the system prompt to try and make it less verbose (saving tokens). And if I'm not misremembering, there were a lot of complaints when they changed the default effort from high to medium.
I was surprised to learn that Sonnet generally has the same tokens per second as Opus
Wow, seems worse even on price/performance than GLM 5.2, which is only 744b parameters.
From the system card: "On CyberGym vulnerability discovery, Claude Sonnet 5 is less capable than Sonnet 4.6, and far less capable than Opus 4.8 and Mythos 5
As with the other evaluations in this section, these results were achieved with all safeguards turned off. When run with our default mitigations, Sonnet 5 scored a 0 on CyberGym"
That's for their `JSON` data types. In DuckDB it's just a string meaning lots of queries will have to do JSON parsing on every row, but the inserts are very fast. Definitely a bit of a footgun and when you actually just need STRUCT or MAP.
There's a talk about ClickHouse's approach from its creator: https://www.youtube.com/watch?v=xHj9mysh0GI , but the gist is that it maintains (sub)columns to store different paths in the JSON
In other ways DuckDB has very good JSON support, like you can do `CREATE TABLE name AS `SELECT * FROM 'data.json';` and it'll infer the schema when possible.
It's great but you definitely pay for it. Encoding can be really slow, and to a lesser extent decoding as well.
So I still end up using .jpg quite often, or .webp as a good middle ground
My favorite spatial reasoning benchmark: https://minebench.ai/
no tricks, I'd definitely be curious to know how much screenshots help
They reserved the option to buy it at this price, and are now exercising it
If the government takes the bulk of your income after a certain point, there isn't really that big of a push to create ground-breaking technology.
I'm skeptical that high taxes is a large reason to lose to California of all places. Maybe in some important sense CA has "earned" that via talent and funding density while NL hasn't (from the perspective of a company, to be clear)
I'm almost certain $2800 is actually too low if you're really hitting weekly limits.
I'm on the $100/m plan and used $300 at API billing yesterday (according to ccusage)
Seems like one session is >$100 and I can get 10 full sessions per week
The $200/m plan is supposed to be 4x that in usage, so with 2 of those you could use 4*2*100*10=$8000 in just a week
Using Simon's numbers here as a bare minimum https://simonwillison.net/2026/May/27/product-market-fit/#en... you'd get 1200*4*2=$9600 a month
If they ban GPUs we can always multiply the matrices on paper
Interesting comparison, thanks for sharing! It reminds me of this post about how machine learning and encryption have some fundamental similarities: https://reiner.org/neural-net-ciphers
I can certainly imagine LLMs taking a similar path. Maybe it's useful to think about what fundamental differences could contribute to LLMs taking a very different path. What comes to mind is the scaling hypothesis, implying that the best LLMs will require enormous capital investment.
That seems largely incompatible with open source barring a fundamental change. There's open weights, but I can't think of a clean historical analogy there and find it extremely difficult to even guess how the future will go
My guess is that they liked the status quo with Project Glasswing and didn't want Fable to be public, especially if anyone is jailbreaking it into Mythos and using it for cyber
But then it backfired spectacularly and now it seems they can't use Mythos currently
I'd guess it's absurdly good business wise, given API pricing is like 10-30x as much
It's certainly worse news for Anthropic than other labs since it's not completely random, and there's people in the administration (e.g. David Sacks) who don't like Anthropic -- perhaps seeing them as an enemy
Seems like estimates are that 70-85% of their revenue comes from API usage/pricing, so some users switching from Opus to Fable for that would've had a big impact
Then there's people switching from GPT 5.5 or upgrading their subscriptions, and Fable being scheduled for removal from subscriptions on the 23rd
Does low/medium effort fix it for you? Seems like Fable 5 low can outperform Opus 4.8 high/xhigh often, and uses a lot fewer tokens
Looking at page 255 of the model card (https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c3...) it might be much better on all dimensions (speed, cost, quality) to just use Fable 5 on low/medium effort than switch to Opus
Looks like it'll be more like $2t
https://polymarket.com/event/anthropic-ipo-closing-market-ca...