@sama Must be a pain in the ass process...
I dont get why the us gov. is blocking cheaper models release like Luna.
good model names btw congratz! Eager to test & integrate them in our pipeline.
@kevmoo Flutter is already incredible to build powerful web apps but advanced text (high volume, MD rendering, extensibility) is still a bit rough and require extra attention to deliver 60FPS+ experience in any case ( but always achievable ).
That + isolate on web would be legendary !
@onixya2@CaroleQuintaine Mais ou es le souci franchement ! les enfants savent apprécier des parents qui ne renoncent pas à leur personnalité, leurs passions et leurs moments de folie en devenant parent.
Préjugés de merde, hypocrisie et étroitesse d'esprit 🤡
@tristanmf not so unpopular opinion je suis d'accord malgré que je sois un pan-enthousiaste (j'accepte le terme croyant). Un très large faisceau d'indices mais pas de preuves indiscutables.
Mais de la à discard l'ensemble du sujet mécaniquement c'est l'autre extrême qui s'illustre.
@patloeber huge gemini supporter since 1.5 but honestly this (in our scenario) would be a real deal breaker. Hope 3.0 preview superbely priced vs intelligence ratio is not going to be ditched. There is a life after strong coding models and many different needs.
@patloeber Hi Patrick what is the status for the flash 3.0 preview ? It get replaced/deprecated in the future by 3.5 (with the 3x+ cost increase) ?
Dont you feel 3.0 is great for many tasks that doesnt require the 3.5 intelligence ?
Or is 3.5 lite is going to be on par with 3.0 preview ?
@KARLOLIVE monter sur le capot comme un gros déglingos ça crée les liens rapidement.
On sent d'ailleur toute la sincérité de votre amitié 😂
L'arène politique un cirque sans pareil !
@ValsAI@OfficialLoganK the problem is that you are retiring models agressively and replacing them with smarters but higher cost (not small increments to say the least) leaving us with no choice except looking elsewhere and creating our own model garden (which is maybe at the end the correct move)
@ValsAI@OfficialLoganK you are right that agentic systems require at the end more intelligence and the way we built things is evolving but imho we still need base-intelligence models for things like context pruning, segmentations, sub tasks before passing that to reasoning/analysis steps adv. steps.
@bindureddy What about the associated costs (x3 + lot of tokens for thinking) ?
I mean from my tests it is super costy. You dont have this issue ?
Of course if you come from Anthropic this must be a relieve 😁
@OfficialLoganK only one question remain: Is google plan to force customers to use flash3.5 x3+++ pricing with no replacement for actual flash3 (or model with same price/intelligence range) ?
If no alternative provided or clear communication i'm sorry but this feels like a trap !
@OfficialLoganK@theo you really should have created a "pro-flash/turbo" model additionally to the og flash 3
The message you give to us is -> great now flash is on par with pro but 3times more expensive so "good luck dear loved customers with the products you built and priced with our models".
@theo exactly that's why i'm moving away from Google.
I was a super engaged early hardcore supporter when no one thrusted google on ai but i feel betrayed by these model huge price changes and the lack of fallbacks they offer. They dont care about crushing your business.
@OfficialLoganK feel ripped by the abusive price change of flash...
Knowing 3.0 flash will be sooner or later deprecated without a proper alternative in this price range i think its time to move on
@OfficialLoganK@GoogleDeepMind you keep saying tokens prices goes down. I agree with that if you integrate the intelligence 3.5 has but you really need to consider offering a flash lite 3.5 at least as good as flash 3 og model
@OfficialLoganK@GoogleDeepMind its a very good model but way too expansive.
A good solution would be to scale the price for token input/output relative to thinking effort.
@sundarpichai amazing model for its size and a good learner !
Make it de-facto accessible also in the gemini api / vertex ai api like the frontiers models with proper quotas.
It fill the gap that flash25 lite deprec leave since 31lite is in another league of intelligence but also price.