wolttam

The timing of this is so awfully strange.

However, this U.S. centric view of the future of AI is wild. If the U.S. prevents its businesses from using open-weight models, only those businesses will suffer, while the rest of the world flourishes with the access to cheap, good-enough, intelligence.

Multiple dollars per million input/output tokens was never sustainable for the majority of use-cases - hardly anybody outside the U.S. can afford that and many within it can’t. Models costing that much will find less reasons to be used over time, not more.

All the while their capabilities will continue to shift towards smaller and much cheaper models, at least until we hit some kind of true data limit with them

Would be great to see Kimi here.

I like Bash for its portability and pretty much no other reason. That one attribute of it is pretty hard to replicate with any other tool, at least over short time horizons

The U.S. model is honestly more baffling to me, if the goal is broad economic growth.

But it hasn’t seemed like the U.S. powers have been interested in broad growth for quite a long time now. Just whatever can line their own pockets.

This is very cool, and the paradigm that scales to bringing truly useful robots to virtually all domains of life - just as LLMs have scaled to that degree across virtually all domains of knowledge work.

I picture a lot of perfectly tended gardens growing food.

DeepSeek isn't the only one publishing; Kimi published about both their Attention Residuals and KDA / sparse attention.

The U.S. has ~350 million people.

Anthropic and OAI can piss and moan all they want - limiting the U.S. to only their models would hurt the U.S. economy in myriad more ways than the failure of a couple of companies that scaled too quickly. If they get that outcome, the rest of the world would simply keep moving forward with access to open models and tokens at pennies on the dollar.

It’s 100 yuan per million output tokens in China. That’s $14.7 USD - not “far cheaper”.

Or, GLM 5.2 simply had more time in the RL oven.

Deepseek V4 Flash, the 284B model, is roughly equivalent to launch GLM 5, the 744B [sic] model.

Nice, they updated it again

Mythos/Fable-class models have been around for at least 4 months internally in the US, and Kimi still isn't quite there, so I'd say the 6-months is still about right.

I know that nobody owes me anything, but I still cheer when people decide to share rather than own.

I'm so glad to be wrong!

It was on the page linked in the top comment, but it's been removed.

https://platform.kimi.ai/docs/guide/kimi-k3-quickstart this one, it used to have more information about the model itself, similar to the K2.6 and K2.7 pages.

Edit: OpenRouter still describes it as an open-weight model: https://openrouter.ai/moonshotai/kimi-k3

Guess we'll see!

It did say that, but it doesn't any longer.

I'm a bit nervous this one isn't going to be open-weights. Any mention of "open" has been struck from the literature for this model (it was present an hour ago). We don't even know active params?

At this pricing, I'll be surprised if it's open.

I guess it remains to be seen whether this will be open-weights. We don't even know how many active params at this point.

I am commenting only to say that I read the reflected-letter text and found that amusing.

This wasn't the LLM, it was Grok CLI preemptively uploading the entire CWD, regardless of where that CWD is, to its own server.

I don't think it is reasonable to expect every user (including those just starting out with the tools - maybe experimenting, maybe younger/less experienced in general) to think that the tool they're running for the very first time is going to automatically exfiltrate all of their data.

It's a pretty serious fuck-up. This guy tweeted about it, who knows how many didn't even notice. It should have been opt-in, it should give user an indication that it's about to do this, etc.

Andrew may be a little bit grating to some (including me at times), but I value the perspectives that he brings.

There is a lot of discussion about the exact tone and phrasing, etc, of Andrew's post. There's something there - we expect perfectly composed writing, never getting emotional, never saying how we actually feel about a specific individual's behaviour. Meanwhile, in private, we often let those emotions fly, name names, etc.

I think Andrew gave us an actual look into how he feels about Jarred -- ambivalent, largely.

This is a truly underrated approach IMO

and on-device model combination right now

That would depend entirely on what your device is. This sounds likely not to be an issue with the harness, but the capabilities of the models you've tried.

I experience almost no tool call failure using my nothing-special harness and DSv4 Flash.

It's impressive indeed. I would also expect the next checkpoint of DSv4 Flash to come in somewhere at this level (DeepSeek has had over 2 months to continue training since it released).

It's exciting that the open models continue to get better and more efficient across the board!

Hy3 lacks the DSv4 architecture's KV Cache efficiency.

Whereas I can run DSv4 Flash on a pair of DGX Sparks and have enough memory left over for 3M tokens of KV cache, with Hy3 (quantized to FP4), there is only room for ~130K tokens of KV cache.

I'm looking ahead to the next wave of open-weight models that are as efficient as DSv4 (which is really efficient), and have been heavily distilled on GLM 5.2 (which is trivial, given it is open weight)

To expand on Chinese models:

- DeepSeek

- GLM (Z.ai)

- Minimax

- Kimi (Moonshot)

- Hy3 (Tencent)

- Qwen (Alibaba)

(Each one of these with weights available to download and run locally)

It's the native DeepSeek V4 Flash quant which is released in MXFP4. You have enough RAM left over with 2 sparks for ~2M tokens of working KV cache.

I was recently using self-hosted DeepSeek V4 Flash to poke around the DSpark implementation in vLLM (well outside of my domain)

I did wonder if I was doing anything Fable would have flagged - sounds like yes.

we pass the PFAS to others

Is there no way to filter them out of withdrawn blood?