Hacker Newsnew | past | comments | ask | show | jobs | submit | small_model's commentslogin

Used it, very fast but tiny context window and doesn't have good reasoning. (good for quick simple code changes)

MIMO 2.5 Pro ultraspeed has a 1M window. 1,000 tok/sec is great for planning since you can have a rapid conversation with a lot of turns.

Agreed, 1000tok/s just fills up the context window (which is big by 2004 standards) super fast. But seems like 5.3-spark was just a taste of what’s to come.

2004 standards? O.o

In 2004, I took a class where we trained "language models" that were bigram word models, on an archive of a couple years of the Wall Street Journal.

I remember someone who literally announced they were dropping the class to the whole room at the end of a lecture, saying "This isn't AI!!!"


1904

Back when we were kids, we would get 0 tokens/sec _if we were lucky_

The suckers are those companies agreeing to this deal. 'Your margin is my opportunity' means prices will fall eventually once more production come on line. The invisible hand will slap their faces

"... involve a commitment to buy a certain quantity of product and pay for it in a pricing band that has a floor and a ceiling price. The floor price covers the historically high gross margins mentioned above, and the ceiling price means those who commit to an SCA are insulated if memory prices go even higher."

So clearly 16 large buyers consider it likely that prices will go even higher. How likely? >10% chance? Likely enough to sign an agreement.


"Eventually" doing a lot of work. Micron (and implicitly anyone signing this deal) are betting demand is going to outstrip capacity for several years, taking into account what new capacity can be brought online and when.

Well, apparently those companies believe memory prices will continue to rise, so they'd better lock in supply at the current (high) prices. We'll see if they're right...

More production can only come online if the producers have money to do so.

Turns out Apple fucked Micron in 2023 so they couldn't expand capacity: https://9to5mac.com/2026/06/25/micron-exec-suggests-apples-a...


nobody is agreeing to what the headline says. the SCAs are just a hedge against even higher prices. the agreements will be broken the second prices drop.

It's a five year agreement. Good luck financing, constructing, equipping, and bringing new memory fabs online in less than 5 years.

Yes, we will get a crippled version of Mythos, 5.6 and future models, while the chosen few will have unfettered access.

It's the year of the open source AI model is the new 'It's the year of the Linux Desktop'. It's not and never will be for 90% of people

That’s not true at all. While not as good as proprietary models they are still very good and can do A LOT, certainly more than their cost would make it seem.

It’s only a matter of time before companies start to acknowledge the huge cost of tokens and look for a cheaper alternative with basic cost-benefit analysis.

My F500 company is getting local infrastructure going to host open models and I’m sure many will just switch to bedrock + the best open models.

It’s foolish for companies to let three companies dictate the price of tokens, I just don’t think they are aware of this now by and large.


GLM 5.2 is competitive with Opus 4.6.

I thought it was better than that? It matches 4.8 in many evals and even beat Fable in Design Arena by a very healthy margin.

Well if us gov would block people from using windows or macos, then it may well be.

AI is a great revolutionary tool for work, but it is still a tool and needs humans to drive it. Obviously companies heard the promise of "Replace your large headcount expense with cheap tokens" and creamed their pants. Its funny to see them walk back, it will be at least a few years if not more before it replaces humans fully (and will need another breakthrough)

Back in dot-com, there used to be a website called f'ed company that chronicled the dot-com dead pool. This time around there needs to be a similar website that records AI walk backs so it helps the mgmt class not make stupid decisions.

Agree, not only code but the plethora of tools usually required in a dev job, Jira, git, GitHub, build agents etc. These can all be accessed by the agent via API/MCP so you can say to the agent, describe my next ticket, create a branch off latest master and investigate the code re the ticket, later, codex/claude review the code (from other model), move the ticket to blah, add a comment, check the build (oh it failed dig into the logs), find any code review issues etc. All from a terminal, this used to be annoying clicking through web pages. I just eyeball the code and make sure it looks fine. Job done.

Does Waymo make a profit yet? This looks like them giving up on being the main robotaxi choice, Teslas will win on scale and cost per mile so they are going for the premium market. Not sure thats big enough to turn into profit.


Teslas also can't drive themselves so I doubt Waymo is worried much about them.


Cant they, tell that to the ones in Austin, Dallas and Houston that are currently taxis people around


dude the waymos in phoenix are dumb asf sometimes, fsd is much smarter situationally


10 years behind Tesla, they are doomed


Actually, 137 years behind Tesla ;)

https://spectrum.ieee.org/axial-flux-motor-yasa


... Eh? No. No current large-scale production car uses axial flux motors. You're talking concept cars, plus a few oddities like https://en.wikipedia.org/wiki/McLaren_Artura


I mean in EV's, a fancy expensive concept motor isn't going to help, autonomy should have been their direction 10 years ago, they are on the downward slope of their existence im afraid. Its the equivalent of Blockbuster adding a tape rewinding service to try and boost sales vs Netflix streaming.


That is also not guaranteed. VW (with its array of brands) leads the european BEV market by a lot. [1] (sorry I only found data for April 2026 right now but earlier months were similar)

One additional point of data. In Q1 of this year they delivered 200K BEV worldwide [2] while Tesla did 350k [3].

Calling that 10 years behind is not warranted in my opinion. I would agree to say competitive and challenged.

[1] https://cleantechnica.com/2026/06/01/europe-ev-sales-report-... [2] https://www.volkswagen-group.com/de/pressemitteilungen/volks... [3] https://ir.tesla.com/press-release/tesla-first-quarter-2026-...


Volume alone is a bad metric, Tesla makes profit on their EV's sales Mercedes makes a loss, not sustainable.


My understanding is that all current Teslas use radial flux motors


The latest frontier models will write code better than you and more elegant, with less lines of code, in 100th of the time, with full test coverage. Hand coding is like writing out assembly/machine code rather than using a compiler.


This hasn’t been my experience. State of the art models available to the public still do all sorts of bandaids and bad hacks. Putting code where it doesn’t belong. Stapling types onto variable (in TypeScript) when abstractions/types already exist to use. I use it to generate code, but still have to review every line and have corrections/steering basically every time.

Maybe you have access to some other model?


This.

Insisting on writing code by hand when LLMs are available is not software engineering in 2026. Engineers find the most cost-effective solution for the problem at hand that meets the requirements.


> The latest frontier models will write code better than you and more elegant

They often do, but they often don’t. I regularly have to push for more elegant, or less lazy solutions.


It's not massive amount of code for the sake of it, it's that it built a complex large piece of software itself. If you compared the linux kernel LOC to some guys toy kernel on GitHub the LOC difference should give you an idea on the complexity. "Yeah why is Tolstoy flexing about his number of pages in his book, I wrote Spot goes to school and its only 20 pages"


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: