[baq]

agreed, the next price increase from frontier labs (and the inevitable limits decrease in subscription tiers) will have people thinking real hard about their model providers and that's when mistral should be ready. however, given their recent performance, I realistically don't have my hopes high up.

[amunozo]

DeepSeek is both cheaper and better than Mistral.

[barrell]

Not in many tasks. I use deepseek as a fallback in https://phrasing.app and it’s always very apparent when it happen (due to mistakes/clear performance drop off)

[efromvt]

Interesting - which models specifically? I'd be interested in using mistral over deepseek if it was competitive (guess I need to go benchmark)

[barrell]

I use small, large, an medium-3.5 depending on the task

[gregorygoc]

Because they distill

[losvedir]

I feel like there's an implication here that distillation is a problem but I don't understand what you mean. I thought distillation was generating text from a model and then training another model on it. Is the something unethical in that? You're paying the API costs to generate the tokens, right?

Or I guess more to the point: is this something frontier labs have said is (or tried to paint at any rate) problematic? This feels like an "out of the loop" situation because I've only ever heard "distillation" with a positive connotation before.

[taneq]

Whether it's a 'problem' or not is viewpoint-dependent but it's against the OpenAI ToU:

> You may not use our Services for any illegal, harmful, or abusive activity. For example, you may not:

> [...]

> * Use Output to develop models that compete with OpenAI.

Source: https://openai.com/policies/row-terms-of-use/

(I'm also curious whether they consider developing a competing model to be illegal, or harmful, or abusive...?)

[Matl]

> it's against the OpenAI ToU

Given that OpenAI doesn't care about training on copyrighted data, why is suddenly their ToU something anyone should care about?

[jononor]

That OpenAI was in the wrong when they ignored everyone copyright, does not make it right to ignore their ToU. If a one wants IP and rule of law (incl contracts) to be respected, one should not violate others rights when it is convenient.

On a more risk-strategy level there is the size of their legal team, general endowment, and supplier and political connections to consider.

Everyone is free to ignore their ToU, but I can understand why a company would avoid it...

[zhivota]

Feels like this should be some kind of anti-competitive violation even if it's not actually. Probably moot under this admin but still.

It's like saying you can't use windows to develop an OS, or drive a Ford on the way to your job at Hyundai.

[CamperBob2]

In heavily-unionized areas, you'd best not drive that Hyundai to your job at the Ford plant.

[ctrlkctrls]

it doesn't matter the reason. This is a race and nobody will care or remember how the winners got there.

Mistral looks like it's fading away to irrelevance unless they can play alongside the similar sized models, or have some unique advantage other than being in Europe, for Europe. I was really excited for them back when they were startup that had the biggest European venture round ever. This space will have a few winners, and many losers. Google, plus either Anthropic or OpenAI most likely. Big models will see breakthroughs in inference performance/cost fall precipitously and small models will only exist on devices (Pixels and iPhones, cars, watches, bluetooth speakers, etc)

[gregorygoc]

It’s not that I don’t agree with you, I am just pointing out why it’s hard to catch up to scaling laws given the European economic (capital) and political (US would be upset if they found out Europeans distill) constraints. China is only bound by economic constraints.

[KronisLV]

With the insight in your comment and this bit from the above one:

> This is a race and nobody will care or remember how the winners got there.

It seems like the EU should have paid China for the distillation datasets, esp. since Mistral isn’t even a governmental org.

[sofixa]

> This is a race and nobody will care or remember how the winners got there.

For consumer AI, yes. For coding assistants, probably.

For specific application "business" AI like the things Airbus announced the other day? Not at all. What matters for an Airbus using Mistral to build compliance documentation based on AI generated physics simulations is the enterprise relationship, reliability, compliance, forward deployed engineers helping with the fine tuning, quality, predictability, support. A Chinese lab having a better at benchmarks model that is cheaper is just irrelevant for that.

And IMO, the real money in AI is this type of "business AI" deployment. Developer tooling tends to converge on becoming commoditised. Once you're a core supplier for a big bank and embedded in their processes, you're there untill you screw up with the pricing (like Broadcom), and even then.

[k__]

Why doesn't Mistral distill?

[gregorygoc]

Good question, given that American companies basically threw copyright law into the trash, I think they should.

[Saline9515]

American companies can't sur Chinese ones, but they can do it with European ones.

[Matl]

So then the European ones should join with European copyright holders to sue OpenAI/Anthropic and watch them trying to BS their way around what they train on.

[cleaning]

American companies won't even sue other American companies (xai distilling opus), in fact they instead form partnerships ;)

[xienze]

Well if they did wouldn't be able to feel superior to Americans about that particular thing. Perish the thought!

[__m]

It’s really a pity, why can’t they feel superior while breaking ToS and copyrights just like Americans can feel superior over Deepseek while breaking ToS and copyrights?

[opsnooperfax]

I suppose losing with dignity is a consolation.

[djvdq]

Also, new Medium 3.5 is far more expensive than previous Mistral models, and much more expensive than e.g. Deepseek

[KronisLV]

I tried it out on some dev tasks with their Mistral Vibe subscription, and the performance was pretty okay (okay, not great), both in regards to development and speed. Worse than Anthropic's models I'm used to but at 20 EUR per month it wasn't a bad deal - except that the 200k context size would more or less be a deal breaker in many cases.

[eisa01]

Where do you sign up for that subscription?

I wanted to try out Mistral, but I fail to find anything like that even after creating an account

[KronisLV]

The other comment already mentioned that you get their subscription: https://mistral.ai/pricing/ they do say that you can try out their coding agent for free, but personally the Pro tier is pretty affordable too to try out for a month.

Then you can install their coding harness, I personally used the Python + uv option: https://mistral.ai/products/vibe/code/ if you don't have uv yet, you might have to install it too: https://docs.astral.sh/uv/ though I already use it for other projects. Oh and if on Windows, you probably want to do all of the installation inside of WSL, just so that file paths are the *nix variety, I've had issues otherwise with pretty much every coding harness, like OpenCode as well (across multiple models).

After that, you need an API key for your subscription, you can generate and copy it here: https://console.mistral.ai/codestral/cli that's also where you see the quota, though it seems to NOT refresh instantly, but more or less a few times a day.

Either way, happy coding!

[djvdq]

Maybe on their pricing page?

https://mistral.ai/pricing/

[bermudi]

Everything is more expensive than deepseek. They aren't frontier in intelligence but they are the frontier in cost per intelligence