For the best experience on desktop, install the Chrome extension to track your reading on news.ycombinator.com
Hacker Newsnew | past | comments | ask | show | jobs | submit | history | sebdufbeau's commentsregister

As a stealth model, it was priced as $1.25M in / $10M out

Right now, it seems free when you are a Cursor Pro user, but I'd love more clarity on how much it will cost (I can't believe it'll be unlimited usage for subscribers)


A bit late but it's actually not free. You can see it on their models page. It's similarly priced to GPT-5 and Gemini 2.5 Pro.

https://cursor.com/docs/models#model-pricing


Couldn't find the info quickly, is the stealth model cheetah they had a few weeks ago their new Composer 1 model? If not, who's was it?

Edit: yes it was: https://x.com/amanrsanger/status/1983581288755032320


Has the API rollout started? It's not available in our org, even if we've been verified for a few months

EDIT: It's out now


it is out yet. i poll the api for the models and update this GitHub hourly.

https://github.com/spullara/models


OpenRouter says Singapore


Zhipu is the most Chinese company out there. Almost all of their fund is from state-owned entities now.


I joined a project last year which uses Konva inside an Angular application. It was my first time doing any kind of canvas programming, but the ergonomics of konva can be picked up quite rapidly and I've been enjoying quite a lot.

We're working on performance-sensitive project, so one lesson we learned is that all shapes listen to all mouse events by default. We didn't even have lots of shapes, but this was enough to have a noticeable performance hit due to all the event handlers being registered. We pivoted to an opt-in approach instead and that fixed most of our problems.


"We're working on performance-sensitive project"

Konva looks awesome, but canvas based. For more performance I switched from canvas to pixi, which is webgl/webgpu based.

Drawing can be also expensive there(in some cases even more so), but if you can manage to put it in a texture in time, you can have looooots of moving animated shapes even on mediocre mobile phones.

https://www.goodboydigital.com/pixijs/bunnymark/


Impressive, it took me 113K bunnies for a single frame drop on an iPhone 14pro


At 105 I started seeing dips while adding, but would level back to 60 during bounce. iPad Air 4.

Is this the new benchmark? “The new iPad can handle 200k bunnies”


true, also declarative approach is amazing for wegbl, you need to draw many lines - pass start/end points in a buffer at once.


I wanted to do something like this, but have not found a nice way, yet. You do this by hand? Meaning writing the shader code yourself?


By hand or regl library (almost by hand). I tried THREE, but it feels it gives me more pain than benefits (and you still end up tweaking their shaders sometimes). AI is pretty good at small tasks like tweaking shaders. I needed to draw a spectroscope, like a line connecting all pixels on the image one by one on the colorspace projection. Ended up 100 times more efficient in webgl comparing to browser API. But it is pretty low level. On another hand once you dig deeper into shaders, they are actually more powerful. Let's just say, API is just painful too :). Also, there is some fun with how transparency works and order of elements displayed (because of parallel processing). Oh, ideally I want web canvas API, but which allows to pass a buffer of lines to draw instead of one ;)


Thanks. I started getting into shaders with WebGPU, but that is not widespread yet and WebGL I did not liked, so I likely will get back to writing WebGPU shaders for better performance, once there is decent support. (And my current solution works, just could be more performant)


An important point in the story: while she was contacted by the scammers when she was still married, it is after the divorce that she received most of the money that was eventually scammed - over €700k


The article is an excerpt from an upcoming book, maybe it's part of it


How are you comparing bundle sizes? I know that the reported size on some sites like Bundlephobia for v5 is bigger than reality:

https://twitter.com/TkDodo/status/1714275314745081895


Thats a good point. I checked using the "better" tool (pkg-size.dev) they show on your link and it seems they are actually comparable for the "default" export (but im not sure what that gives me with tanstack query - its effectively everything with SWR).


For me, Spotify on Mobile has a sleep timer functionality and one of the options of "End of episode". I do have premium so cannot confirm if it exists on the free tier.


I don't have any hard info, but here's my speculation:

The Spotify most people know (finding and playing songs, creating playlists, recommendations), which obviously very important, is probably a tiny fraction of the complexity for the company.

I would wager the vast majority of the employees at Spotify are working on everything else needed to enable the above. The artists portal, analytics and royalties, paying the labels, paying the artists, legal for every country, etc.

Someone who knows could probably elaborate more than me on everything needed to run a company like that.

Does that require 8000 people? Debatable, but it's certainly more complex than clicking a song and playing it.

EDIT: A quick glimpse at job listing for Spotify on LinkedIn shows:

* Client Partner in India

* Business Operations in Nigeria

* Creator Partner Manager in Mexico

* Artist & Partnerships Manager in Egypt

* NLP Research Scientist in the USA

* Director of Sales in Singapore


Well you are right - Spotify operates in 184 markets, so even if you only had a small 10 person team for each country handling sales/marketing/support/localisation, that's already c2000 employees before you start on anything else.

Remember Spotify has approx. the same number of premium subscribers as Netflix has subscribers, so it's not exactly tiny.


Monaco, Syria, North Korea, Lithuania don't need 10 person teams. It's this napkin math that causes tech companies to inflate. I'm sure Spotify execs did the same math and told HR "Hire 10 people per country, thx".

In the Age of AI, even before 2023 and ChatGPT, localization is a very automatable+contractor heavy job that is very easy to complete. All of africa might need 3-4 teams of 20 people total, and that's including sales/marketing. Maybe double for South America.

The reason tech companies have expanded so much is it looks better when an 8000 employee company controls the worlds music than when an 800 person company does. Same with Google, Facebook, AirBnB, Netflix, etc. The tech monopolies want to make they monopolies seem less ridiculous to regulators.


80 people to manage African countries from a finance, sales, marketing, support, legal, localisation and licensing perspective for a company the scale of Spotify is way too light.

Localisation isn’t just about language - consider payment requirements, legal and regulatory specific challenges, pricing strategy, regional licensing, there might be specific features required to support low bandwidth (eg carrier agreements), if you want to ad-support like Spotify you will need a team to sell ads… it’s hardly a chatgpt + few contractors job.


> monopolies seem less ridiculous to regulators.

No.

Over-hiring is probably the same everywhere. When "times are good", everyone in the organisation wants to hire because their number of subordinates makes them seem more important and accomplished.


Spotify makes money from record labels who pay good sums for a placement on various ”hot new metal”-lists. Managing those deals takes lots of personnel


Include in that list: ingesting content (audio and video), maintaining clients for <N> platforms, strategic partners support. Algorithmical personalisation is also quite important for Spotify's offering, etc.


Consider applying for YC's Summer 2026 batch! Applications are open till May 4

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

Search:

HN For You