Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- Comparing compute cost versus FlashAttention-2 is not very honest to me.
FlashAttention-2 is not used anymore for at least 2y.
This architecture would have been a massive improvement 3 years ago, but it is a ~solved~ problem IMO.
by samber - There was, let's say, significant skepticism the last time they announced something. What's changed?by chrsw
- I have no idea if the evaluator themselves is trustworthy, but it was supposedly independently evaluated by Appen: https://www.appen.com/whitepapers/benchmarking-subquadratics...by supern0va
- by EDM115
- > SubQ 1.1 Small scores near-perfect at 1M, 2M, 6M, and 12M tokens. The model was trained predominantly at 1M tokens yet the retrieval held near perfectly at 12x that length, despite compressing attention to just 0.13% of relationships. This generalization is a direct consequence of SSA routing attention based on content relevance rather than fixed positional patterns.
If the results persists from 1M to 12M, why not 24M or 48M? Sounds almost too good to be true.
With back of the napkin math from inside my head, that'd be like 0.5/1 million LOC, depending on language/code density, could just fold the entire codebase into one prompt if it's a small one, that'd be neat :)
- It likely falls off very steeply after that. 8 to 1 (which I am assuming based on the 0.13% figure) is a pretty common ratio for sparse matrix stuff.
- Interesting idea but until I get my grubby little fingers in it, to try it - difficult to have an opinion.
I am hopefully expectant that we will see all sorts of optimizations in the next few years that will enable even more local model use and slash commercial API costs. I get excited by the results when I enjoy one or two short coding sessions a week with Claude Opus but it is even more exciting to get a major task done and see that I only used $0.05 for DeepSeek v4 Flash or perhaps $0.15 for DeepSeek v4 Pro. It was exciting in even a different way when I two shotted a complete TypeScript/Tauri app using gemma-12b-qat with little-coder on a cheap laptop a few days ago.
- It's been all talk and no action ever since their first announcement.by satyarohith
- It's easy[1] to promise, it's hard to deliver. I hope the best for them.
[1]: https://magic.dev/blog/ltm-1 (note the date)
by kristjansson - According to Subquadratic, Needle in a Haystack is strong up to 12m tokens, but RULER has not been tested above 128k tokens ??by samber
- we need some better standard long-context benchmarks.
needle in a haystack is not good for this, yes it proves the model can attend to its context, but in its usual form, somewhat trivializes the query-key relationship.
something like long-form Q&A would be more ideal. Like reading a book and answering questions that require synthesizing information derived from either the whole thing or disparate portions of it. Like describing an entire character arc in a 1000 page novel with examples and evidential moments.
by bthornbury - > SSA replaces the O(n²) dense attention pass with a learned sparse formulation that scales linearly with context length.
> At 1M tokens, SubQ 1.1 Small requires 64.5x less compute than dense attention and runs 56x faster than FlashAttention-2.
Awesome stuff. Solving context at the model architecture layer rather than trying to bolt on extra memory is the right direction IMO.
by wxw - This one's interesting, and I think the next frontier for LLMs should really just be, how can we get something like Opus 4.6 to cost drastically less, for the same output? I say 4.6 because from 4.6 onwards it's been pretty darn good, at least for me, always feels like every model upgrade someone hates it, heck even 4.5 was fine.
- I agree on opus 4.5-4.8, but Fable 5 was a noticeable upgrade.by mritchie712
- Yes - I want that and dramatically faster. Newer models don't seem to need any more or less guidance and iteration, so let's make the time-to-wrong-answer as short as possible.by robmccoll
- I don’t understand why this lab is allergic to providing details on what they actually made, especially when Chinese labs are more than willing to share architectural specs/code/kernels (eg NSA/FSA, RAMBa, HISA, DSA LightningIndexer, etc). I don’t doubt that they’ve done something here, but the lack of details makes me default not trust this, particularly when this is the second time that they’ve released a “technical report” that just waxes poetic about the concept.by cmogni1
- You don't understand why the thing their entire company is valued upon is...not being given away freely? They literally are taking an open source model and then adapting it with this technique. If they disclose it, the frontier labs will immediately copy it and outperform them.
My guess is that they're angling for an acquisition.
by supern0va - Well, I know this is possible because I have built things that work just like it is promising to do. The two key technologies needed are:
- guided window attn. Predict where to attend to but in a fixed window. If you do this to just the token/vocab you can keep effectively unlimited context and perfect recall. (yes, I can do that. There is a trick to teaching it how to predict position. This also immediately opens other crazy things like NN memory)
-efficient fixed state size models. So not a recurrent mechanism because that breaks training, parallelizable like transformers, but fixed sized state instead of unbounded attn. Pick a reasonable amount of state and it is amazingly good since it doesn't need to keep separating wheat fro chaff in context (yes, it is possible to build this, I have. It works. This also opens up real streamed models. I have a true infinite context streamed model I toy with locally that I am getting to be audio/text in and audio/text out in real time.)
Put those together and you have O(1) token gen, infinite context and perfect recall. It is a whole new world of models. You can interact with a model until you have it at the state you want and then save its state and use that as if it were your system prompt. Batches pack perfectly so inference is massively more efficient. Training is massively more efficient. Transformer and unlimited attn models are a dead end. But how do you make money on this as an independent researcher? If I release the Two Weird Tricks this is all based on I get zip and the big players get even more tech for free. If I keep it all secret I get Zip and eventually the tricks will be figured out. (Yes a little frustration here) If anyone wants the model architecture of the future make me an offer :)
by jmward01 - >A full breakdown of the mechanism and how it compares to FlashAttention, DeepSeek sparse attention, and recurrent architectures is in the Technical Report.
Oh they did publish details lets read the technical report!
> The mechanism by which SSA meets these requirements is outside the scope of this report
TFGs...
by kristjansson - Business wise, it would make sense to hold off on details till they're at least ready to serve. Look at what happened with Open AI and reasoning models. Everyone struggled with getting RL to work with LLMs for a good while. Open AI figured it out, and a few months later everyone had their prototypes out in short order. Don't forget who these labs employ. They're some of the brightest people around. Sub-q aren't really in a position for that lol. If they'd shared details at the first announcement for instance, the big labs might have had something out by now while they're still pulling resources to scale and then what ?
- They don't need to provide any details at all. They just need to give people access to their model and charge them for it. That they don't do that and instead pay for external evaluations indicates that they believe people would be unimpressed if they could access the model directly. The only purpose of this press release seems to be making investors give them more money.by yorwba