Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Gemini 3.5 Flash

blog.google

Post reply on HN
Print view View original

671–680 of 692 posts

Re: Gemini 3.5 Flash

#671

Post by easygenes »

WarmWash wrote: post #632
DCKing wrote: post #583

Earlier quoted context omitted.

If two things hold up - 1) this is actually a 2-300B parameter model and 2) this is actually competitive with frontier OpenAI and Anthropic models (and not just benchmaxing), the implications are pretty big. It would mean you could run "frontier level" performance in one box at home. 300B models at least fit in a single maxed out Mac Studio or a small stack of DGX Sparks or AMD Strix Halo boxes. For comparison, DeepS…

Opus is estimated to be around 4T parameters, and 5.5 around 9T. [1] And while 3.5 at least qualifies to be in the same neighborhood, which is stunning if these numbers are all true, it may be that closing that last ~10% difference needs 50x more parameters. [1] https://arxiv.org/pdf/2604.24827

Their methods are only calibrated on open models (of course) and they admit very broad confidence bounds. You can also just see from comparing their estimates of the same models at different reasoning levels that there are major confounders to this. I would err on the absolute lowest side of their estimates for frontier models (e.g. 3T for GPT-5.5, 1.5-2T for Opus 4.5+).

Re: Gemini 3.5 Flash

#672

Post by sheeshkebab »

akulbe wrote: post #512
sheeshkebab wrote: post #417

Earlier quoted context omitted.

You can only run heavily quantized models on all 3/4/5 rtx gpus (with 32gb or less vram) - and you probably want moe versions like Qwen 35b for this to run at speed somewhat comparable to Claude. It’s still not there to be honest but getting there. Personally I mess around with llama.cpp on m5 max with 128gb - it’s a decent setup to try various medium sized things, and runs llms surprisingly well without quantization…

How is that machine for local inference? It's a serious consideration for me, but getting to hear more from folks that already have it would be helpful.

It’s great laptop to mess around with llms, it won’t replace claude opus or even sonnet.

Re: Gemini 3.5 Flash

#673

Post by nl »

stymaar wrote: post #610
nl wrote: post #542

Earlier quoted context omitted.

Elon says Opus is 5T (and I would expect he'd know) > It's not that frontier labs can't create a 5T+ parameter model, but they don't have the data to optimize a model of that size. The have plenty if data. They use very large amounts of verifiable synthetic data in (lots in coding and math) cover the gap. Also the frontier labs are paying people to do tasks, tracking the trajectories and training on that. Most of the…

> Elon says Opus is 5T (and I would expect he'd know) Even if he knew, why would anyone expect Elon not to lie about anything? > The have plenty if data. I don't think data is the problem either, but compute is: if you want to train your 5T params model like modern small models are being trained (with a thousands time more training tokens than params), that's an enormous training run.

> if you want to train your 5T params model like modern small models are being trained (with a thousands time more training tokens than params), that's an enormous training run.

Yes it is. Spending $100M on training runs is common, and $1B might be in scope for some of the large models.

Sonnet 3.5 cost "a few 10s of millions of dollars" back in 2024: https://simonwillison.net/2025/Jan/29/on-deepseek-and-export...

Re: Gemini 3.5 Flash

#674

Post by LarsDu88 »

DCKing wrote: post #583
easygenes wrote: post #466

For those who would like to know the total and active parameter count of this model: even though Google doesn't disclose the model technicals, we can infer them within relatively tight margins based on what we do know. We know they serve the model on TPU 8i, which we have plenty of hard specs for (so we know the key constraints: total memory and bandwidth and compute flops). We can also set a ceiling on the compute c…

If two things hold up - 1) this is actually a 2-300B parameter model and 2) this is actually competitive with frontier OpenAI and Anthropic models (and not just benchmaxing), the implications are pretty big. It would mean you could run "frontier level" performance in one box at home. 300B models at least fit in a single maxed out Mac Studio or a small stack of DGX Sparks or AMD Strix Halo boxes. For comparison, DeepS…

TurboQuant. They can fit more in less now

Re: Gemini 3.5 Flash

#675

Post by DCKing »

LarsDu88 wrote: post #674
DCKing wrote: post #583

Earlier quoted context omitted.

If two things hold up - 1) this is actually a 2-300B parameter model and 2) this is actually competitive with frontier OpenAI and Anthropic models (and not just benchmaxing), the implications are pretty big. It would mean you could run "frontier level" performance in one box at home. 300B models at least fit in a single maxed out Mac Studio or a small stack of DGX Sparks or AMD Strix Halo boxes. For comparison, DeepS…

TurboQuant. They can fit more in less now

TurboQuant is a runtime optimization for a model's KV cache and doesn't allow for reduction in model size.

Re: Gemini 3.5 Flash

#676

Post by forsalebypwner »

WarmWash wrote: post #57
aliljet wrote: post #33

Is there a good benchmark tracking hallucinations? The models are all incredibly good now, even the open ones, and my hope is that the rate of hallucinations is something that's falling off in concert with larger and larger context lengths.

People complain about them incessantly, but I can almost never get people to actually post receipts. Every provider allows sharing chats, and anyone can share a prompt that reliably produces hallucinations. More often than not, people are using images in responses that go awry. Which is fair, the models are sold as multi-modal, but image analyses is still at gpt-4.0 text-analyses levels. Also knowledge cutoff issues,…

https://gemini.google.com/share/dbb8e26ca469 from today

Re: Gemini 3.5 Flash

#677

Post by btheunissen »

forsalebypwner wrote: post #676
WarmWash wrote: post #57

Earlier quoted context omitted.

People complain about them incessantly, but I can almost never get people to actually post receipts. Every provider allows sharing chats, and anyone can share a prompt that reliably produces hallucinations. More often than not, people are using images in responses that go awry. Which is fair, the models are sold as multi-modal, but image analyses is still at gpt-4.0 text-analyses levels. Also knowledge cutoff issues,…

https://gemini.google.com/share/dbb8e26ca469 from today

holy shit it was possessed by Richard Stallman

Re: Gemini 3.5 Flash

#678

Post by lmazgon »

marknutter wrote: post #635
lmazgon wrote: post #567

Click on "Listen to article", make sure the voice is "Umbriel" and skip to 4:15 - there's a hallucinated part at the end in Russian (I think). On a blog post about the latest and greatest AI model. Oh the irony.

Looks like they removed the option to "listen to article". I wonder why.

Here's the link to the raw audio file, for anyone who's still interested: https://storage.googleapis.com/gweb-uniblog-publish-prod/med...

Re: Gemini 3.5 Flash

#679

Post by stymaar »

nl wrote: post #670
stymaar wrote: post #610

Earlier quoted context omitted.

> Elon says Opus is 5T (and I would expect he'd know) Even if he knew, why would anyone expect Elon not to lie about anything? > The have plenty if data. I don't think data is the problem either, but compute is: if you want to train your 5T params model like modern small models are being trained (with a thousands time more training tokens than params), that's an enormous training run.

I mean in general I'm pretty doubtful about things he says, but in this he was comparing Grok and it sort of makes sense in the context: https://x.com/elonmusk/status/2042123561666855235

In that context specifically, why would you trust him not to lie?

He's using a massive number for Opus to make Grok look good “for its size”.

If he said something praising Anthropic and like “Grok is 7T, while Opus is better while being only 5T, we need to work harder” or something then maybe I could believe it. But here it's a context where he has all the incentives to inflate Opus' size to make himself look somehow “in the race” when he really isn't despite the money and compute advantage.

Given this tweet I wouldn't be surprises if Grok was actually 1T and Opus being in the same ballpark.

And I'm absolutely not buying current-days Sonnet being a 1T parameters model (that's an absolutely deranged take: that would make Anthropic already behind Chinese model makers, which I think isn't something anyone would put money on).

Re: Gemini 3.5 Flash

#680

Post by time0ut »

I ran through the eval loop for a side project’s task (personalization of a micro video game, no thinking) last night. Head to head with Gemini 3 Flash Preview, results came out at basically a wash on my rubric. The output quality was good, well grounded, and reliable across 144 runs. But not noticeably better. It isn’t a traditional coding task, so can’t infer anything there. The amazing part was how fast it is. It was consistently about 2x faster than 3 Flash Preview and slightly faster than 3.1 Flash Lite Preview which is amazing. For my task, the price difference doesn’t matter, so easy upgrade. I plan to write up a quick blog post with the results over the weekend.
Post reply on HN