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

Mistral 3 family of models released

mistral.ai

Post reply on HN
Print view View original

21–30 of 243 posts

Re: Mistral 3 family of models released

#21

Post by NitpickLawyer »

simgt wrote: post #4

I still don't understand what the incentive is for releasing genuinely good model weights. What makes sense however is OpenAI releasing a somewhat generic model like gpt-oss that games the benchmarks just for PR. Or some Chinese companies doing the same to cut the ground from under the feet of American big tech. Are we really hopeful we'll still get decent open weights models in the future?

> gpt-oss that games the benchmarks just for PR.

gpt-oss is killing the ongoing AIME3 competition on kaggle. They're using a hidden, new set of problems, IMO level, handcrafted to be "AI hardened". And gpt-oss submissions are at ~33/50 right now, two weeks into the competition. The benchmarks (at least for math) were not gamed at all. They are really good at math.

Re: Mistral 3 family of models released

#22

Post by constantcrying »

crimsoneer wrote: post #10
constantcrying wrote: post #6

Earlier quoted context omitted.

The lack of the comparison (which absolutely was done), tells you exactly what you need to know.

If someone is using these models, they probably can't or won't use the existing SOTA models, so not sure how useful those comparisons actually are. "Here is a benchmark that makes us look bad from a model you can't use on a task you won't be undertaking" isn't actually helpful (and definitely not in a press release).

Completely agree, that there are legitimate reasons to prefer comparison to e.g. deepeek models. But that doesn't change my point, we both agree that the comparisons would be extremely unfavorable.

Re: Mistral 3 family of models released

#23

Post by GaggiX »

ot wrote: post #20
GaggiX wrote: post #17

Earlier quoted context omitted.

London is not part of Europe anymore since Brexit /s

Is it so hard for people to understand that Europe is a continent, EU is a federation of European countries, and the two are not the same?

I think you missed the joke

Re: Mistral 3 family of models released

#24

Post by memming »

talliman wrote: post #13
simgt wrote: post #4

I still don't understand what the incentive is for releasing genuinely good model weights. What makes sense however is OpenAI releasing a somewhat generic model like gpt-oss that games the benchmarks just for PR. Or some Chinese companies doing the same to cut the ground from under the feet of American big tech. Are we really hopeful we'll still get decent open weights models in the future?

Until there is a sustainable, profitable and moat-building business model for generative AI, the competition is not to have the best proprietary model, but rather to raise the most VC money to be well positioned when that business model does arise. Releasing a near stat-of-the-art open model instanly catapults companies to a valuation of several billion dollars, making it possible raise money to acquire GPUs and trai…

It’s funny how future money drive the world. Fortunately it’s fueling progress this time around.

Re: Mistral 3 family of models released

#25

Post by mirekrusin »

simgt wrote: post #4

I still don't understand what the incentive is for releasing genuinely good model weights. What makes sense however is OpenAI releasing a somewhat generic model like gpt-oss that games the benchmarks just for PR. Or some Chinese companies doing the same to cut the ground from under the feet of American big tech. Are we really hopeful we'll still get decent open weights models in the future?

Because there is no money in making them closed.

Open weight means secondary sales channels like their fine tuning service for enterprises [0].

They can't compete with large proprietary providers but they can erode and potentially collapse them.

Open weights and research builds on itself advancing its participants creating environment that has a shot at proprietary services.

Transparency, control, privacy, cost etc. do matter to people and corporations.

[0] https://mistral.ai/solutions/custom-model-training

Re: Mistral 3 family of models released

#26

Post by usrnm »

ot wrote: post #20
GaggiX wrote: post #17

Earlier quoted context omitted.

London is not part of Europe anymore since Brexit /s

Is it so hard for people to understand that Europe is a continent, EU is a federation of European countries, and the two are not the same?

Europe isn't even a continent and has no real definition (none that would make any sense, anyway), so the whole thing is confusing by design

Re: Mistral 3 family of models released

#27

Post by p2detar »

sebzim4500 wrote: post #15
yvoschaap wrote: post #5

Upvoting for Europe's best efforts.

That's unfair to Europe. A bunch of AI work is done in London (Deepmind is based here for a start)

That's ok. How could they know that there are companies like Aleph Alpha, Helsing or the famous DeepL. European companies are not that vocal, but that doesn't mean they aren't making progress in the field.

edit: typos

Re: Mistral 3 family of models released

#28

Post by mirekrusin »

talliman wrote: post #13
simgt wrote: post #4

I still don't understand what the incentive is for releasing genuinely good model weights. What makes sense however is OpenAI releasing a somewhat generic model like gpt-oss that games the benchmarks just for PR. Or some Chinese companies doing the same to cut the ground from under the feet of American big tech. Are we really hopeful we'll still get decent open weights models in the future?

Until there is a sustainable, profitable and moat-building business model for generative AI, the competition is not to have the best proprietary model, but rather to raise the most VC money to be well positioned when that business model does arise. Releasing a near stat-of-the-art open model instanly catapults companies to a valuation of several billion dollars, making it possible raise money to acquire GPUs and trai…

Explained well in this documentary [0].

[0] https://www.youtube.com/watch?v=BzAdXyPYKQo

Re: Mistral 3 family of models released

#29

Post by Youden »

timpera wrote: post #2

Extremely cool! I just wish they would also include comparisons to SOTA models from OpenAI, Google, and Anthropic in the press release, so it's easier to know how it fares in the grand scheme of things.

They mentioned LMArena, you can get the results for that here: https://lmarena.ai/leaderboard/text

Mistral Large 3 is ranked 28, behind all the other major SOTA models. The delta between Mistral and the leader is only 1418 vs. 1491 though. I *think* that means the difference is relatively small.

Post reply on HN