Birbla

Join the discussion

Write your take first — we'll ask for email only when you're ready to publish.

Sign in
  • Hacker News
  • Google seems to have anorexia when it comes to model intelligence. They have an internal hard constraint on price per token it seems, and they are trying to squeeze out intelligence with limited compute.

    I wonder if there is something with their TPU cycles that makes them want to postpone training a new model. My guess is that they have been on the same base model for 6 months and they may have waited for the next gen TPUs to train Gemini 4, which greatly limits how much intelligence they can increase and forces them to do cost efficiency increases.

    by spyckie2
  • Google Cloud is probably Google Deepminds biggest competitor. Big company kinda bullshit.
    by WarmWash
  • I'd guess they did model-hardware codesign but the design ended up limiting the scaling capability of the model (i.e. they overoptimized too soon).
    by logicchains
  • It's probably a mix of things but I do think they are viewing "edge AI" as their strategic play: on-device, small efficient models (Android / iOS) and instant AI summaries in google search etc. So all of their focus is on delivering strong performance in a compute constrained environment.

    I do think it's still also simultaneously true that they have an actual problem with competing with current frontier progress. It's just that has gone from an existential threat to something they are willing to defer addressing because they see the long game for them sitting at the smaller end.

    by zmmmmm
  • Could it be that they have to serve their models to billions of users?
    by JacobAsmuth
  • Wow - Google does not even bother to show benchmarks of these models compared to the frontier and Chinese labs - only against previous versions. I'm not surprised. Having worked there for years it was amazing just how inwardly looking the company is.
    by velominati
  • No word about updating Jules, which is still stuck on 3.1 Pro. I get that it's probably niche but I've really appreciated basically being able to give directions to Jules on my phone, then reviewing and merging a GitHub PR fifteen minutes later. It's been great for getting some progress in on a few personal projects during my commute when I can't exactly pull out my laptop.

    Anyone have any good alternatives?

    by b473a
  • Claude Code
    by haberdasher
  • If you own a Raspberry Pi or similar: Hermes + Tailscale + iSH over tmux.
    by steven_pareto
  • I have no affiliations with the team or product, but Superconductor reminded me of Jules when I tried it a couple of months ago.

    It might be overkill features-wise, but there's a free tier and it likely won't be left for dead anytime soon.

    by christoff12
  • I do not, however I am curious about Jules support. I didn't know if this was a dead project or not. Seemed really interesting but then I didn't see much development/announcements/discussions around it. Last update from their changelog was as you said 3.1-pro support in March.
    by bespokedevelopr
  • Both Claude and Codex can code in the cloud, it works quite well!

    I tested Jules and while the idea is good in theory, I found the model's intelligence to be very lackluster.

    by aweb
  • It's scary relying on Google's models.

    I have a very price sensitive workload that used to run on flash 2.5 lite - it's deprecated now.

    The replacement 3.1 flash lite is a lot more expensive, but now also has a sunset date.

    3.5 flash lite is even more expensive.

    So the price is rising and you have no choice but to keep paying more and more.

    by swe_dima
  • > So the price is rising and you have no choice but to keep paying more and more.

    I presume you can't use deepseek?

    by raducu
  • Just switch the model, its not that much effort tbh. And u can also get a cheaper model than 2.5 lite for the same intelligence
    by hagen8
  • same I just switched to OpenAI after using flash 2.5 lite for almost everything at our company. We spent thousands just to build this workflow now Google says screw off
    by zuzululu
  • And somehow, the most annoying is not even the price hike, but it is that is you expect to build a product on any of theirs models, they spend their time being deprecated and you have like to be on the lookup to start from scratch selecting a model and fitting it every year or so... Impossible to have any stability...
    by greatgib
  • I felt the same way about openai's text-davinci-002 and code-davinci-002 (gpt-3.5). They were amazing completion models and openai basically dumped them with no equal cost or equal performance replacement. Instead all their models are opaque with no ability to work in completion mode where one actually controls the text input to the model.

    These days no company even has completion models where one controls the text input fully. Worthless.

    by superkuh
  • I'm running price-sensitive data extraction workloads on flash 2.5 and its still the king when it comes to accuracy + cost, all the gemini 3 variants perform a bit worse and cost a lot more. Low-key freaking out, ngl
    by pdntspa
  • That’s part of why, since Firebase, I’ve tried to never depend on Google products for business, especially not GCP.

    Features stay in Beta for ages, whatever that actually means, and released ones get deprecated things fast.

    Where some of the competitions treats deprecating entire services as "let’s not put it on your frontpage, put deprecation notices all over the doc, and politely ask new users not to start new project with them".

    by ElFitz
  • I moved directly from 2.5 flash lite to deepseek v4 flash, its already cheaper and if your prompt caching is good you can save so much more money.
    by rayboy1995
  • It is both less intelligent and more expensive than GLM-5.2, while being closed weight.
    by jgbuddy
  • Is that statement based on token price? More and more it seems that $/token hides as much as it reveals. Token efficiency, tokenizer differences, etc. I'm not saying that you are wrong, I am just saying it is becoming a bit more difficult making statements like this without a bit more research.
    by SubiculumCode
  • It’s multimodal though.
    by dyauspitr
  • It's roughly equal on price and intelligence as GLM 5.2 while being ~8x faster.
    by jbellis
  • It's also about 15x faster.
    by JacobAsmuth
  • It'd be interesting to know how much the Intelligence as a Service angle serves as a value-add in the minds of Google's executives.

    You can get decent open-weight models now. That's not difficult. The difficulty is 1) running them and 2) compliance.

    My company runs Claude on GCP's Vertex AI solution. We're in the US healthcare IT space, so the models need to be from somewhere that American healthcare agencies and companies have traditionally been okay with sourcing code from - which means the US, Canada, and maybe Europe. The stuff that handles PHI/PII must be in the US. The expense of hosting is more of a PITA than most customers want to go through this early in the technology's lifecycle, and intelligence gains are simply a matter of degree for most business tasks.

    In theory, we could find some open-weight model (likely from China) for our development agentic work and host it anywhere you can host AI models. We don't, though, and I think Google, OpenAI/Microsoft, and Anthropic see that as the core of their business.

    by lenerdenator
  • Google's models are always very well spoken and much more pleasant to talk to. And I've used the open weight models a lot, too.
    by kzrdude
  • It's also got vision and audio. So, the better comparison is any of the other large Chinese open models that are better and cheaper than Gemini Flash.
    by SwellJoe
  • But they make up for it by shipping it late.
    by drob518