跳到正文
Hacker News:AI 熱帖·· 2026-10-06AI 評分61

Mistral Large 4「Le Chonk」推出,討論聚焦基準與價格

Mistral Large 4: "Le Chonk"

AI 導讀

Mistral Large 4(「Le Chonk」)推出後,Hacker News 討論聚焦基準比較、定價和權重開放安排。有留言稱權重目前尚未開放,承諾月底前開源;另有留言指出它仍落後於中國開源模型,並批評基準圖表的縱軸未從零起算。

正文

I don't know why, but I personally find Mistral's marketing strategy much more appealing than that of other companies.

For example, there's something about Anthropic's picked design and their little Claude avatars that's unsettling to me.


Sans context, I really like the Anthropic and Claude's faux-academic, minimal-ish, intellectual-ish branding.

(Especially the way it looked ~12 months ago -- it's gotten more cluttered since then. Perhaps unavoidably, as the breadth of their offerings has grown)

But over time it's begun to feel like unsettling cognitive dissonance as their ambitions grow and the stuff to worry about has piled up.


> faux-academic, minimal-ish, intellectual-ish branding

By this you sourely don't mean the messages shown in Claude Code, where Pi would show "Working..."

Academic style: Cooking... Sautéeing... Julienning... and similar annoying faux-brogrammer moody status messages.


The entire Anthropic branding is religious kitsch - deeply off putting, but apparently quite reflective of their reality.


I think there's such a thing as throwing too many marketing and sales people and too much polishing and "refinement" at something. I see in new product announcements from Microsoft as well. It's like seeing someone try too hard to impress you.


I do my best to avoid any AI marketing because I extremely despise it. I just haven't quite decided on my new profession, yet, but it's either going to be with plants or with animals.


Very nice progress. Also I respect putting Kimi on those charts. Regardless of if they are beating the Pareto frontier (not now), model diversity is a good thing for humanity — I’m hopeful for the team to keep increasing their gains.


It's French, duh. They'll probably surrender half of the company to the Germans after receiving a strongly (compound-)worded letter.


I'm literally English and I feel the need to defend the French here, they're among the most successful military powers in the world historically speaking. It's hardly fair to judge a thousand years of French military history by the outcome of one war where they didn't do well.


It's a joke, chonky is used to refer to fat cats, and there was a joke meme over the summer that Mistral are going to release a new model, Le Chaton Fat (chaton is kitten in French). The name is a nod to the memes.


Can't wait until Mistral starts adopting fancy product line names.

Soon we'll have Mistral 6 Chaton, Mistral 6 Guépard, Mistral 6 Tigre, Mistral 6 Dents-de-sabre, Mistral 6 Beast King, etc.


Is the reason for the massive gains in certain benchmarks due to distillation from the other lead models hence the slightly "under" pattern seen in the comparison charts?


The main thing I always get away from the comparison tables of these "big" models, is how well Deepseek v4.1 Flash performs. While still being the cheapest model by a long shot.


Excited to try this. The low costs v. benchmarks alone here are worth a serious test. K3 has been my daily driver for a month or two now and it's dramatically reduced token spend (while not having much of a negative impact on productivity).

This was the era of the AI race I was waiting for.


Haven't tried it yet. It looks like sol is a hair more expensive on input, a hair less expensive on output ($2/in, $10/out per 1m, K3 = $0.82/in, $13/out per 1m).


> Trained from scratch

How are they training without pirating the Z library corpus and all that?


Don't believe Mistral. They're wrong. It's really called "Le chaton fat".

Also 1T-A49B. Weights currently closed but promise to open source them by the end of the month.

Great release movie.


>Don't believe Mistral. They're wrong. It's really called "Le chaton fat".

OpenAI's therapist: Le Chaton Fat isn't real and cannot hurt you

Le Chaton Fat:


Bar charts should start at zero. If they don't start at zero, there should be a clear visual indicator that the chart has been trimmed without having to read the axis labels. I hate that this has to be repeated so often that it has become a cliché.


A bit disappointing to see it still lagging behind Chinese open models. Those Chinese models are pushing proprietary models to raise the bar, but we need equally strong non-Chinese open models to challenge the Chinese ones in turn.

來源:Hacker News:AI 熱帖 · news.ycombinator.com