Rendered at 08:15:55 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
irthomasthomas 11 hours ago [-]
Insane. 3 points behind opus on the artificialanalysis index.
Mimo cost ~$400 at the old price, so about $40 today.
Opus cost ~$5000
That's over 100x cheaper, and just 3 points behind.
I can't wait to experiment with an llm consortium of 100 deepseek and mimo models.
Crazy times.
Shut up and take my m̶o̶n̶e̶y̶ data!
Edit: Gemini on google search told me I could write strikethrough text on hn using <s>. Mimo told me it was unsupported and then went on to list some tags that are supported, like <b>bold</b>. I tried copy pasting the word in strikethrough from a word processor but it lost the format. I ended up using mimo in an agent shell wrapper to produce it, and copy pasting from the terminal worked for some reason.
_davide_ 51 minutes ago [-]
I had a subscription before the price was cut down; the model kept randomly looping the with same character (burning 30% of the budget in one shot), and the overall performance for agentic purposes is, simply put, terrible.
It finds non-existing bugs and randomly removes chunks of code to fix them, then even presents it as an "extra fix".
Maybe it's a good generalistic model; I haven't tested it in that regard.
MiniMax (currently 2.7) which is a ~270B model tuned exclusively for agentic purposes, performs so MUCH better; it's more reliable and cheaper. Both are still far away from Opus 4.7 that I'm using at work. IMO benchmarks are just a very rough estimation; everyone cheats as much as they can get away with. Test the model yourself; do not make any assumptions based on the benchmarks.
I would love to see specialized, cheaper, bleeding-edge models like MiniMax for other non-agentic purposes as well. Why pay $1 for a general model when, for example, you can pay $0.1 for a content-moderator model that you actually need?
maxdo 5 hours ago [-]
benchmarks we deserve: google search quick ai answers vs full llm model :)
noman-land 11 hours ago [-]
What did MiMo say?
irthomasthomas 11 hours ago [-]
Says its not supported and lists a few tags that are, like <b>bold</b>
Does this work: s̶t̶r̶i̶k̶e̶t̶h̶r̶o̶u̶g̶h̶
noman-land 7 hours ago [-]
Well done. Unicode wins again. 𓂺
wg0 13 hours ago [-]
That's deliberate. US AI companies have no chance of recouping even fraction of their valuations.
PS: Have not tried this but Deepseev4 Flash (not even Deepseekv4 Pro version) with set to "high" has pretty much Claud Opus 4.7 level of capabilities and is lightening fast and dirty cheap. Hours and hours of conversation barely costs few cents.
bel8 13 hours ago [-]
DeepSeek Flash on high (not max) is a freak of nature indeed.
Very disproportionate intelligence-to-cost ratio.
I'm leveraging this temporary anomaly and using it as my coding workhorse.
fgonzag 11 hours ago [-]
The weights are open and when prices settle down again will be runnable with less than 10k of hardware.
I can easily run it in a 8 bit quant with the 4 x 48GB Radeon Pro W7900 GPUs I snagged for 2k each before the memory squeeze.
A 158B parameter model, especially in an architecture as efficient as DS4 is not that hard to drive currently if you got in before the craze, and will be relatively easy to drive with future hardware generations.
UncleOxidant 9 hours ago [-]
Doesn't DeepSeek-V4-Flash have 284B parameters?
fgonzag 7 hours ago [-]
You are correct.
binary132 12 hours ago [-]
what makes you so confident it’s temporary
giwook 6 hours ago [-]
> US AI companies have no chance of recouping even fraction of their valuations.
A big caveat here is that many US companies (particularly in sensitive industries, like defense) will likely not want to (or not be allowed to) use Chinese models for anything of substance.
syntaxing 5 hours ago [-]
What about self host Chinese models?
monster_truck 5 hours ago [-]
Still no.
elboru 5 hours ago [-]
Why?
Haven880 7 minutes ago [-]
USA is as censored as what we believe about China. You will get cancelled by Americans if you use "communist" stuff. That is why hardly any Chinese EVs in USA. Because it is communists stuff. The odd thing is iphone is made in China. So it is more of selective enforcement when convenience. Chinese AI even self host means your will influence by communism. You want McCarthy era back again?
carimura 5 hours ago [-]
trust, optics, laws
Arcuru 13 hours ago [-]
I am very happy with DSv4 for their price/performance but neither of them are comparable to Opus.
chillfox 5 hours ago [-]
Yeah, I really like and use DSv4Pro for personal projects, but I also use Opus all the time at work and they are definitively not at the same level.
I can only conclude that people who claim they are aren't doing anything close to the edge of what these models are capable of or any niche things.
I would say DSv4Pro is around the same level as Sonnet.
tartoran 13 hours ago [-]
But they're overall a good thing for us consumers even if we'll never use these models, it forces the prices down all around.
surgical_fire 13 hours ago [-]
I have been using DeepSeek API within Claude Code. So far it has been legitimately superior to Claude, and Codex that I used before.
giwook 6 hours ago [-]
Anecdotal evidence is nice but hard to take seriously given the myriad variables at play here.
NitpickLawyer 13 hours ago [-]
Since the 3rd party providers on openrouter have all converged on much higher prices in serving these models (both mimo and dsv4), there's obviously a question on how/why are they lowering the prices so much.
It's possible they've finally integrated cheap(er) chinese chips. It's also possible they're just subsidising inference for real-world usage data. Interesting either way.
Aurornis 13 hours ago [-]
> there's obviously a question on how/why are they lowering the prices so much.
Same reason they release some of the models for free: They are trying to capture market share.
lordofgibbons 13 hours ago [-]
The difference is that releasing the model for free doesn't have ongoing cost for the company. Providing cheap tokens is very expensive - specially if you don't have access to the latest transistor node chips. So I think the parent comment is right, there's something else at play allowing DS and Xiaomi to offer these nearly free tokens.
jack_pp 5 hours ago [-]
LLM providers can't "capture" anything. People loved Claude Code because it was cheap and good. Not cheap anymore? People switching to Codex, DS4 etc.
Their only moat is maybe being SOTA but that only lasts so long before everyone else catches up.
lucaspiller 5 hours ago [-]
This is why they are pushing more for non-tech folks to use their products with desktop apps. They are not going to switch on a whim.
Bolwin 5 hours ago [-]
I mean there is a minor moat. Most people don't enjoy switching providers or models. If you can get people to trust you'll stay near frontier, they'll stick around even when you aren't the best. Claude is a prime example of this
HDBaseT 3 hours ago [-]
I switch models all the time.
/model in OpenCode
There is no "moat" for me.
Using the standard chat applications as a normal conversational/question has a little bit of moat as its able to cross reference existing conversations, but I disable that mostly anyways to prevent as much data retention as possible.
zrn900 13 hours ago [-]
> how/why are they lowering the prices so much
Like I responded to someone else:
- Cheap electricity
- Cheap, domestically produced GPUs
- Efficiency research. (a lot of it from Deepseek's research)
Also, the Chinese government wants the AI to be as accessible as EVs so everyone will use it.
slake 2 minutes ago [-]
Also if this is on the path of anything the Chinese do in the physical goods world, inference will be rockbottom cheap in a few years because they'll invest in the hell out of energy, GPUs, research, etc. The same thing they did with EVs.
Only artificial barriers will keep people using some of the frontier stuff in a couple of years. No costs will justify.
baq 13 hours ago [-]
National security, training data
merelydev 8 hours ago [-]
[dead]
passive 15 hours ago [-]
I worked part time with MiMo 2.5-pro over the last month, and barely managed to use 500 Million of the 700 Million tokens I had allocated.
My plan was just upgraded to 38 BILLION tokens per month. That's at least 10X the tokens I've used in my entire agentic development so far.
I should probably downgrade my plan, but we'll see. :)
6ak74rfy 5 hours ago [-]
Token allocation/cost aside, how was the quality of the model? Any comparison with any other model you've used?
For example, I've heard DeepSeek v4 Pro is comparable to Sonnet 4.7, so I just bought some credits to try it out.
sisve 13 hours ago [-]
Did you not get 38B units? And a token = 2.5 unit (cache hit) or up to 600 unis (cache miss)
passive 12 hours ago [-]
Yeah, I think they did switch the unit type.
zrn900 13 hours ago [-]
Yep. I also got stupefied after I logged in and saw how many tokens they stuffed into my account...
CachedaCodes 14 hours ago [-]
These and the Deepseek ones that were were cost reduced recently are perfectly capable models for the vast majority of light work and more.
It's funny thinking the US companies are hiking prices and Chinese ones do the opposite, it's obviously an strategy, but pretty funny
MaxPock 13 hours ago [-]
How are these "capacity constrained" Chinese companies running inference without Hoppers and Blackwells ?
lukax 13 hours ago [-]
Huawei Ascend AI Accellerators. DeepSeek V4 model architecture was optimized for Chinese hardware.
martinald 13 hours ago [-]
They can (not entirely sure how 'grey' market this is) either have subsidiaries outside of china (eg: singapore) that provide the inference and/or just rent it off the public gpu clouds.
dijit 13 hours ago [-]
Making their own NPUs for inference probably, you don't have to buy NVidia for inference. Google doesn't.
interesting/funny: their off-peak rates apply 00:00-08:00 Beijing time, so nine-to-five for someone on the NA west coast :p
dubcanada 13 hours ago [-]
China has a population of 1.4B, US is 349M. 0-8 Beijing time is their off-peak? How is that funny, that's literally how timezones work?
bel8 13 hours ago [-]
It's funny, in a good way, because their off-peak times match perfectly the werstern peak demand.
ericmay 6 hours ago [-]
Can folks in China run US-based models? Seems like they should take advantage of this overlap in peak timing.
linzhangrun 6 hours ago [-]
Yes, use VPN; they are the main clients
ericmay 5 hours ago [-]
Why do they use a VPN?
esseph 5 hours ago [-]
Chinese state firewall?
weird-eye-issue 4 hours ago [-]
Well not just that, OpenAI explicitly blocks them
hootz 13 hours ago [-]
Will try MiMo now. I have been mainly using just DeepSeek lately because of the fact that V4-Flash destroys basic work for basically 0 cost. Haven't exceeded even 50% of my OpenCode Go weekly limits using V4 Flash and Pro.
sourcecodeplz 8 hours ago [-]
You can use Codex as an orchestrator and claude code via mimo/deepseek api as executor. I've read this a lot before but when you really try it, it is really something in the way you can stretch your credits.
827a 6 hours ago [-]
Hot take: The reason this is happening is because the market for Chinese AI models hosted by Chinese companies is struggling. Even the market for Chinese AI models hosted by western companies is soft: During the week of May 18, OpenRouter processed 3.4T DeepSeek v3 Flash tokens (their most popular model). Google has announced that Gemini is processing 746T per week; Claude is probably processing more. And the Chinese models were already staggeringly cheap, far cheaper than most Gemini, Claude, or GPT models, before this recent array of pricing changes.
Broadly: No one is using the Chinese AI models. Everyone, globally, everywhere, including in China, is using the models from OpenAI, Anthropic, and Google. The models from the Big Three western labs represent >80% of all tokens processed and likely >95% of all revenue.
runako 5 hours ago [-]
> OpenRouter processed 3.4T DeepSeek v3 Flash
> Gemini is processing 746T per week
I read this totally differently. A startup nobody really knows is doing half a percent of Google on a commodity task?!? Google, which puts Gemini on billions of devices by default, without the user asking? Google, which is distributing Gemini to users who are unaware they are even using it?
Versus a startup that does not even have a login button on its homepage?
This is astonishing.
827a 5 hours ago [-]
Unfortunately, the market doesn't generally let you buy Blackwells with "we got half a percent of Google's marketshare with a model we're literally giving away for free [1]". You need that thing we call Capital. But, they may certainly opt to have it written on their gravestone, as Google is (checks notes) continuing to put Gemini on billions of devices and doing quadrillions of tokens per month.
This is a bizarre comment for a couple of reasons.
First, obviously everyone involved understands that someone has to pay to provide a free service. Everyone involved also knows that this sometimes makes sense as a business strategy (I have not paid to ship anything from Amazon for close to two decades).
Second, OpenRouter's business model specifically does not require them to run all (any?) of the models available through the platform. Provider is one of the choices when you choose a model, and each provider can have separate pricing.
The link you posted shows only one provider, Crucible. That may/may not be affiliated with OpenRouter? Even assuming an affiliation, it's opaque who is subsidizing this usage. Is it OpenRouter or Crucible?
All of this is somewhat of a distraction. Even if someone gave search away for free (like Google), it would still be an accomplishment to get to half a percent of Google's volume. Or to sell half a percent of the volume of Android phones. Or whatever.
Kudos to the OpenRouter team!
827a 4 hours ago [-]
In the statement "we got half a percent of Google's marketshare with a model we're literally giving away for free" the term "we're" here refers to the conglomeration of "DeepSeek" (for making a model small enough to be capable of being hosted for free) and the model providers who do offer it for free (why they do this is... unknowable). It does not refer to OpenRouter, who are merely middlemen.
My original DeepSeek v4 Flash token counts spanned all providers of that model, both paid and free; I merely pointed out the free provider to substantiate a point that DeepSeek's product may be so bad that they could quite literally give it away and people would still prefer to pay (a lot) to OpenAI, Anthropic, or Google. Why this is the case, I leave as a exercise to the reader; I'm just citing numbers and facts.
HDBaseT 3 hours ago [-]
Agreed.
Not to mention, week on week more and more tokens are being processed via OpenRouter. [0]. The number keeps going up, with no end in sight in my opinion, if the China models continue offering cheaper inference, whilst tailing behind not too far, the line will keep going up.
OpenRouter is not the only "router" type AI company. More fixed providers like OpenCode and commandcode are offering subscription services on open/china models, likely consuming billions of tokens each. Who know how many tokens are being process directly against Deekseek and Kimi's APIs.
gitah 4 hours ago [-]
If you're going to compare OpenRouter numbers for DeepSeek at least use the same metric to compare Gemini. During last week DeepSeek V4 Flash did 3.72T tokens which is way higher than combined token counts for Gemini (2.5 Flash + 3.5 Flash + 3.1 Pro)
DeepSeek's official API, which has 10x cheaper cached input cost isn't even on OpenRouter as a provider, so just like Google, most volume is not going through OpenRouter. (Gemini's official hosted api is on OpenRouter BTW)
Also you're comparing an API with Google's internal corporate and consumer app use. Bytedance announced they were using 63T tokens/day (441T / week) at the end of 2025, so they are probably even higher than Google now. We don't know how much weekly tokens the DeepSeek chatapp uses, but it would also be a very high number much higher than OpenRouter tokens.
For the real reason of the recent price drops, go ask your AI about how much it would cost to run DeepSeek V4 or MiMo 2.5 after Ascend 950 PR have started to be mass delivered in 2026 Apr at $10k / card.
Bolwin 5 hours ago [-]
OpenRouter is not indicative of volume. Most high volume clients will go to the providers directly. There's not point to paying the 5% OR cut if you know what you want.
827a 5 hours ago [-]
That's just it: This is not happening with the Chinese models, because western corporations are the primary drivers of AI adoption globally and western corporations are not signing up for a DeepSeek API key. If they're working with Chinese models at all, which they rarely are, it is via a western-hosted provider like Bedrock, Vertex, or OpenRouter; or self-hosting. Sure, hobbyists and individual programmers might be comfortable forming a business relationship with a nationalized Chinese entity, but you'd need a microscope to see that relative to the spend that, say, Eli Lilly is throwing at Anthropic every week.
But you're right that OpenRouter is only one data point. It is, unfortunately, one of the few we have.
chillfox 4 hours ago [-]
That makes some sense.
I mean, I am going to use the best I can afford. And at work that's Opus, but while work is happy to let me spend $50+/day, that's just not viable for personal hobby use, I need to keep that in the realm of a WOW/mmo subscription.
skeledrew 6 hours ago [-]
These reductions as Microsoft and Uber say AI is too expensive. The play is right there.
admiralrohan 7 hours ago [-]
The token plan is confusing.
From their docs "After using 10M input (cache miss) tokens of MiMo-V2.5-Pro, it is equivalent to consuming 3000M Credits, and you can still enjoy 1100M Credits of MiMo-V2.5". So it's around 12M input credit vs Earlier 60M tokens.
Flockster 13 hours ago [-]
The 99% is with regards to cached inputs. It seems to now at the same price as deepseek v4-pro
sim04ful 13 hours ago [-]
"The api pricing for mimo-v2-pro and mimo-v2-omni remain unchanged" could we presume this means the discount isn't from hardware improvement or availability ?
zrn900 13 hours ago [-]
VSCode + Cline + Mimo v2.5 pro works ! great !. Give it a try.
nh43215rgb 13 hours ago [-]
So exactly same as deepseek 4 api pricing
bel8 12 hours ago [-]
One difference is that MiMo 2.5 (non-Pro) has image, audio and video input capabilities.
DeepSeek does not understand image, audio or video.
indigodaddy 6 hours ago [-]
I've heard non-Pro isn't nearly as good for coding as Pro?
UncleOxidant 8 hours ago [-]
Is this deepseek v4 api pricing for May (at 75% off - offer supposedly end June 1)? Or the non-discounted api pricing?
Deepseek made the discounted price permanent before this.
prodigycorp 13 hours ago [-]
How realistic is this:
Chinese models incidentally slurps up some terms that lead them to finding unflattering words that you wrote about the CCP in a random journal entry, or maybe a social media csv export. You go to China one day and are denied entry due to what you said.
Realistic or no? (yes i know the us is getting bad in re. to what you write online as well)
Models hosted in China are a siren call that I don't feel bad about resisting.
dubcanada 13 hours ago [-]
This statement makes no sense, because you literally said the "US is getting bad". We already gave up all of our data, if you wrote something about the CCP you should already expect they know about it.
znpy 2 hours ago [-]
Besides that, the us govt already has all your data and yet people are criticising it all around, in the open. They can, without repercussions, because the us is a free country.
Chinese people can’t really do the same.
adrian_b 13 hours ago [-]
This may be true about any models hosted by others than you.
At least the Xiaomi models are open weights and you can host them yourself, avoiding such concerns.
csomar 13 hours ago [-]
Well, at least for the Chinese models you can run them locally vs. the US models that requires you to go through their servers. But to answer your question:
> How realistic is this:
Completely unrealistic unless you are a high value target (journalist, spy, business man, etc...)
> CBP denies travelers entry because of anti-Trump comments
axus 13 hours ago [-]
China won't deny entry for anti-Trump comments, guess I'll use MiMo
artnanika 13 hours ago [-]
China also won't deny entry for anti-Israel comments, so even more reason to use MiMo.
artnanika 13 hours ago [-]
You're projecting the US doing this with criticism of Trump and Israel on China, when there's no proof of China ever doing something like this.
admiralrohan 13 hours ago [-]
Everyone already said what I wanted to say. That all US companies (OpenAI, Anthropic, Google, MS Copilot) have increased price recently while Chinese companies (Deepseek, Xiaomi) are reducing price.
The question is how they are managing to do so? They are supposed to struggle due to chip sanctions.
Secondly, why now? The US companies were supposed to subsidize too but now they are unable to keep up. Everyone going to usage based pricing, so it's unsustainable for them. They are well funded too.
If there are genuine hardware breakthrough reducing compute needs then that is good for the whole world I believe.
vessenes 6 hours ago [-]
> They are supposed to struggle due to chip sanctions
As Jensen has been pointing out for almost a year now, these sanctions were ineffective and probably had the opposite effect of the desired goal.
The history is fairly long, but an inflection point could likely be traced to Trump v1 era DOJ enforcement on (among others) Huawei's CFO Meng Wanzhou in 2018. Huawei was hit with the (really big) stick in international transactions: OFAC violation accusations, and it was a seminal moment in the company's internal operations -- they concluded they needed a fully internal supply chain in China, and retooled for it. Meng Wanzhou cases in the US were eventually dismissed, but she was on house arrest in Canada through 2021 or so.
Fast forward to 2024 -- Huawei was culturally and technically ready to build AI accelerators -- one of the externalities of the sanctions was to provide additional benefit to Chinese companies for buying from Huawei; those economics seem to have provided a boost to on-shore development.
tartoran 13 hours ago [-]
Competition I guess, they must be burning some resources to make this price reduction happen...
lostmsu 12 hours ago [-]
The state of the art models (mostly GPT 5.5, but also Gemini and Claude) are better so they cost more. Qwen 3.7 Max is their only direct competition and it is not any cheaper.
surgical_fire 9 hours ago [-]
Are they?
I have been using DeepSeek, and I am finding it better than Claude or Codex, to be honest.
I don't see myself going back.
c0rruptbytes 6 hours ago [-]
I love ds4, us models are better imo, but like 5% not 500% better, so the valuation doesn't really make sense
that being said, deepseek v4 needs to be on amazon bedrock to actually be feasible in the US Enterprise market and start driving other provider prices down
dzonga 11 hours ago [-]
as someone from the 3rd world - this is pleasant - even 3rd world countries will have affordable "A.I" access via Chinese models.
as someone who now lives & has lived in the west for the majority of their adult life - yeah the US western models r fucked n the crazy valuations of the A.I labs - which also filters down to the economy - since all money instead of being put to productive use is being wasted on this shit. hell electricity bills are up - cz datacenters need power. the current crooks in power don't believe in clean energy.
catlikesshrimp 6 hours ago [-]
>>as someone from the 3rd world - this is pleasant - even 3rd world countries will have affordable "A.I"
I stopped tagging my country as developing and then third world and call it for what it is, a POOR country. I know with increasing certainty that my country will be poor for the rest of my life. I also expect AI to be as available as computers: there are the "have", and there are the "don't have", which is almost always a lifetime condition.
lostmsu 12 hours ago [-]
The price cut is 50%.
sea-gold 7 hours ago [-]
Can you explain any further?
lostmsu 6 hours ago [-]
If you open the page you will see what is reduced by how much.
m3kw9 13 hours ago [-]
Everyone adding "Permanent" to price cuts now
hootz 13 hours ago [-]
Can't be mistaken for someone like, ugh... Anthropic and OpenAI...
dyauspitr 13 hours ago [-]
This sort of pressure will force them to though.
pizzly 6 hours ago [-]
nah, just ban using Chinese models and ban open source models. This will allow them to keep the high price. Got to recoup the money spent somehow, time to lobby the government.
hootz 13 hours ago [-]
I hope so, but I don't know if they are in a position where they can offer these kinds of prices. They are already struggling with not losing a lot of money with their models, while chinese models can be independently hosted by inference providers at a profit already. We need to drive these prices down so AI doesn't become a thing for the few who can pay for expensive subscriptions.
surgical_fire 9 hours ago [-]
They can't afford it. OpenAI and Anthropic bleed money and are desperate for an IPO, that they can get some extra mileage.
rvz 15 hours ago [-]
First Deepseek, Now Xiaomi. A price cut of 99%.
This is why Anthropic wants these chinese AI models banned as they are in the lead in the AI race to zero and they know that there is no modal moat.
So don't tell Dario.
han1 15 hours ago [-]
Like I said. China doesn't care about money. We want AI in people's hands.
culi 13 hours ago [-]
I mean the AI companies probably just want to make American model pricing look ridiculous in comparison (it's working imo). I think the government probably wants actually-useful AI that could be put into chips and actually revolutionize factory work or mining or whatever. Large, SOTA models are not gonna change factory work but extremely efficient and optimized models may
Every industry-wide scale technological revolution has happened because government funded a technology and then opened it up to the masses. Just look at your iPhone: GPS, the internet, AI voice assistants, touchscreens, microprocessors, lithium-ion batteries, etc all came from gov't research (I'm counting Bell Labs' gov't mandated monopoly + research funding as gov't)
Economist Mariana Mazzucato wrote a great book about this called The Entrepreneurial State: Debunking Public vs. Private Sector Myths
zrn900 13 hours ago [-]
> I mean the AI companies probably just want to make American model pricing look ridiculous in comparison (it's working imo)
I really don't think China cares about that. Chinese government's governance logic is making everything so cheap that everyone can get and use it. They did it with EVs and other things. Now they are doing it with the AI.
_davide_ 13 hours ago [-]
They do want to see the American bubble burst, this is the quickest way
andrekandre 9 hours ago [-]
with all the price increases in everything
else, i think we are all tired of this bubble to be honest...
rjhy2020 15 hours ago [-]
OK. Google was just killed.
How is it possible to reduce the price by 99%???????
This is crazy
kingstnap 9 hours ago [-]
The reduction is in cached inputs. I've commented about this before but many labs, except Deepseek and Xaomi now, absolutely scam you for cached reads.
You are basically paying out the nose for a few seconds of VRAM residence if you are giving significant money for cache reads.
The very nature of autoregressive language modeling is that every single output token produced "reads" the cache.
So in principle the price floor for a cache hit is the flat cost of 1 output token.
Now in reality it has to be more than that because you are occupying VRAM with the cache that forces out other users. But it can still be really cheap.
Bolwin 5 hours ago [-]
No one is producing one output token though.
And using up gpus for that cache is a pretty big opportunity cost. I highly doubt it's done in vram. That would be insane for the one hour caches.
So its memory + the time it takes to unload/load into vram + the extra cost per output token
Is it a scam? Idk
sourcecodeplz 7 hours ago [-]
I've read on X that deepseek api can stay alive for hours vs 5 minutes tops for other providers. they do it with ram and ssd, not only vram.
14 hours ago [-]
zrn900 13 hours ago [-]
- Cheap electricity
- Cheap, domestically produced GPUs
- Efficiency research by many phDs. (many AI companies used Deepseek's research though)
dubcanada 13 hours ago [-]
Industrial Chinese electricity costs is similar to that of Texas, It's 8-9cents a kWh. The only benefit is industrial China decides to put millions of solar panels down, so "peak" sunlight hours can drop electricity costs significantly since their rates are highly dynamic.
greenavocado 13 hours ago [-]
Add to that home made inference chips and dirt cheap RAM from CXMT
dyauspitr 13 hours ago [-]
State backed loss leaders.
x3ro 13 hours ago [-]
Is that worse than VC-backed loss leaders? :)
dyauspitr 6 hours ago [-]
Yes. VCs can’t compete with the second largest economy on the planet.
drcongo 11 hours ago [-]
I think this is probably correct based on the way state investment into the Chinese EV market has been working - fund a whole bunch of them and let them fight it out to be one of the few brands that will have the longevity. It's pretty brutal with the cars.
andrekandre 9 hours ago [-]
> let them fight it out
yep, from what i hear, the govt makes sure there is intense local competition in the market so it produces a few really good companies that survive... its kind ironic considering what is going on with mono/oligopolies over here...
readthenotes1 13 hours ago [-]
The rest of the best of the business is paying for it
h4kunamata 5 hours ago [-]
Anything to destroy US tech companies is welcome.
They aren't aiming companies but users which many have no common sense and grant these agentic AI access to everything.
All the restrictions the US imposed to CH, will be reverted back and it will be even worse, because now the data is not reaching the US gov ( we all know they have access to US big techs data ) but CH.
I really hope this goes viral and breaks Nvidia/OpenAI.
ffitch 5 hours ago [-]
fun fact, CH is an ISO code of Switzerland, and China is CN
weird-eye-issue 4 hours ago [-]
I even know the reasoning but CH for Switzerland always bothers me
Mimo cost ~$400 at the old price, so about $40 today. Opus cost ~$5000
That's over 100x cheaper, and just 3 points behind.
I can't wait to experiment with an llm consortium of 100 deepseek and mimo models. Crazy times.
Shut up and take my m̶o̶n̶e̶y̶ data!
Edit: Gemini on google search told me I could write strikethrough text on hn using <s>. Mimo told me it was unsupported and then went on to list some tags that are supported, like <b>bold</b>. I tried copy pasting the word in strikethrough from a word processor but it lost the format. I ended up using mimo in an agent shell wrapper to produce it, and copy pasting from the terminal worked for some reason.
MiniMax (currently 2.7) which is a ~270B model tuned exclusively for agentic purposes, performs so MUCH better; it's more reliable and cheaper. Both are still far away from Opus 4.7 that I'm using at work. IMO benchmarks are just a very rough estimation; everyone cheats as much as they can get away with. Test the model yourself; do not make any assumptions based on the benchmarks.
I would love to see specialized, cheaper, bleeding-edge models like MiniMax for other non-agentic purposes as well. Why pay $1 for a general model when, for example, you can pay $0.1 for a content-moderator model that you actually need?
Does this work: s̶t̶r̶i̶k̶e̶t̶h̶r̶o̶u̶g̶h̶
PS: Have not tried this but Deepseev4 Flash (not even Deepseekv4 Pro version) with set to "high" has pretty much Claud Opus 4.7 level of capabilities and is lightening fast and dirty cheap. Hours and hours of conversation barely costs few cents.
Very disproportionate intelligence-to-cost ratio.
I'm leveraging this temporary anomaly and using it as my coding workhorse.
I can easily run it in a 8 bit quant with the 4 x 48GB Radeon Pro W7900 GPUs I snagged for 2k each before the memory squeeze.
A 158B parameter model, especially in an architecture as efficient as DS4 is not that hard to drive currently if you got in before the craze, and will be relatively easy to drive with future hardware generations.
A big caveat here is that many US companies (particularly in sensitive industries, like defense) will likely not want to (or not be allowed to) use Chinese models for anything of substance.
I can only conclude that people who claim they are aren't doing anything close to the edge of what these models are capable of or any niche things.
I would say DSv4Pro is around the same level as Sonnet.
It's possible they've finally integrated cheap(er) chinese chips. It's also possible they're just subsidising inference for real-world usage data. Interesting either way.
Same reason they release some of the models for free: They are trying to capture market share.
Their only moat is maybe being SOTA but that only lasts so long before everyone else catches up.
/model in OpenCode
There is no "moat" for me. Using the standard chat applications as a normal conversational/question has a little bit of moat as its able to cross reference existing conversations, but I disable that mostly anyways to prevent as much data retention as possible.
Like I responded to someone else:
- Cheap electricity - Cheap, domestically produced GPUs - Efficiency research. (a lot of it from Deepseek's research)
Also, the Chinese government wants the AI to be as accessible as EVs so everyone will use it.
Only artificial barriers will keep people using some of the frontier stuff in a couple of years. No costs will justify.
My plan was just upgraded to 38 BILLION tokens per month. That's at least 10X the tokens I've used in my entire agentic development so far.
I should probably downgrade my plan, but we'll see. :)
For example, I've heard DeepSeek v4 Pro is comparable to Sonnet 4.7, so I just bought some credits to try it out.
It's funny thinking the US companies are hiking prices and Chinese ones do the opposite, it's obviously an strategy, but pretty funny
Broadly: No one is using the Chinese AI models. Everyone, globally, everywhere, including in China, is using the models from OpenAI, Anthropic, and Google. The models from the Big Three western labs represent >80% of all tokens processed and likely >95% of all revenue.
> Gemini is processing 746T per week
I read this totally differently. A startup nobody really knows is doing half a percent of Google on a commodity task?!? Google, which puts Gemini on billions of devices by default, without the user asking? Google, which is distributing Gemini to users who are unaware they are even using it?
Versus a startup that does not even have a login button on its homepage?
This is astonishing.
[1] https://openrouter.ai/deepseek/deepseek-v4-flash:free
First, obviously everyone involved understands that someone has to pay to provide a free service. Everyone involved also knows that this sometimes makes sense as a business strategy (I have not paid to ship anything from Amazon for close to two decades).
Second, OpenRouter's business model specifically does not require them to run all (any?) of the models available through the platform. Provider is one of the choices when you choose a model, and each provider can have separate pricing.
The link you posted shows only one provider, Crucible. That may/may not be affiliated with OpenRouter? Even assuming an affiliation, it's opaque who is subsidizing this usage. Is it OpenRouter or Crucible?
All of this is somewhat of a distraction. Even if someone gave search away for free (like Google), it would still be an accomplishment to get to half a percent of Google's volume. Or to sell half a percent of the volume of Android phones. Or whatever.
Kudos to the OpenRouter team!
My original DeepSeek v4 Flash token counts spanned all providers of that model, both paid and free; I merely pointed out the free provider to substantiate a point that DeepSeek's product may be so bad that they could quite literally give it away and people would still prefer to pay (a lot) to OpenAI, Anthropic, or Google. Why this is the case, I leave as a exercise to the reader; I'm just citing numbers and facts.
Not to mention, week on week more and more tokens are being processed via OpenRouter. [0]. The number keeps going up, with no end in sight in my opinion, if the China models continue offering cheaper inference, whilst tailing behind not too far, the line will keep going up.
[0] - https://openrouter.ai/rankings
OpenRouter is not the only "router" type AI company. More fixed providers like OpenCode and commandcode are offering subscription services on open/china models, likely consuming billions of tokens each. Who know how many tokens are being process directly against Deekseek and Kimi's APIs.
DeepSeek's official API, which has 10x cheaper cached input cost isn't even on OpenRouter as a provider, so just like Google, most volume is not going through OpenRouter. (Gemini's official hosted api is on OpenRouter BTW)
Also you're comparing an API with Google's internal corporate and consumer app use. Bytedance announced they were using 63T tokens/day (441T / week) at the end of 2025, so they are probably even higher than Google now. We don't know how much weekly tokens the DeepSeek chatapp uses, but it would also be a very high number much higher than OpenRouter tokens.
For the real reason of the recent price drops, go ask your AI about how much it would cost to run DeepSeek V4 or MiMo 2.5 after Ascend 950 PR have started to be mass delivered in 2026 Apr at $10k / card.
But you're right that OpenRouter is only one data point. It is, unfortunately, one of the few we have.
I mean, I am going to use the best I can afford. And at work that's Opus, but while work is happy to let me spend $50+/day, that's just not viable for personal hobby use, I need to keep that in the realm of a WOW/mmo subscription.
From their docs "After using 10M input (cache miss) tokens of MiMo-V2.5-Pro, it is equivalent to consuming 3000M Credits, and you can still enjoy 1100M Credits of MiMo-V2.5". So it's around 12M input credit vs Earlier 60M tokens.
DeepSeek does not understand image, audio or video.
Deepseek made the discounted price permanent before this.
Chinese models incidentally slurps up some terms that lead them to finding unflattering words that you wrote about the CCP in a random journal entry, or maybe a social media csv export. You go to China one day and are denied entry due to what you said.
Realistic or no? (yes i know the us is getting bad in re. to what you write online as well)
Models hosted in China are a siren call that I don't feel bad about resisting.
Chinese people can’t really do the same.
At least the Xiaomi models are open weights and you can host them yourself, avoiding such concerns.
> How realistic is this:
Completely unrealistic unless you are a high value target (journalist, spy, business man, etc...)
> CBP denies travelers entry because of anti-Trump comments
The question is how they are managing to do so? They are supposed to struggle due to chip sanctions.
Secondly, why now? The US companies were supposed to subsidize too but now they are unable to keep up. Everyone going to usage based pricing, so it's unsustainable for them. They are well funded too.
If there are genuine hardware breakthrough reducing compute needs then that is good for the whole world I believe.
As Jensen has been pointing out for almost a year now, these sanctions were ineffective and probably had the opposite effect of the desired goal.
The history is fairly long, but an inflection point could likely be traced to Trump v1 era DOJ enforcement on (among others) Huawei's CFO Meng Wanzhou in 2018. Huawei was hit with the (really big) stick in international transactions: OFAC violation accusations, and it was a seminal moment in the company's internal operations -- they concluded they needed a fully internal supply chain in China, and retooled for it. Meng Wanzhou cases in the US were eventually dismissed, but she was on house arrest in Canada through 2021 or so.
Fast forward to 2024 -- Huawei was culturally and technically ready to build AI accelerators -- one of the externalities of the sanctions was to provide additional benefit to Chinese companies for buying from Huawei; those economics seem to have provided a boost to on-shore development.
I have been using DeepSeek, and I am finding it better than Claude or Codex, to be honest.
I don't see myself going back.
that being said, deepseek v4 needs to be on amazon bedrock to actually be feasible in the US Enterprise market and start driving other provider prices down
as someone who now lives & has lived in the west for the majority of their adult life - yeah the US western models r fucked n the crazy valuations of the A.I labs - which also filters down to the economy - since all money instead of being put to productive use is being wasted on this shit. hell electricity bills are up - cz datacenters need power. the current crooks in power don't believe in clean energy.
I stopped tagging my country as developing and then third world and call it for what it is, a POOR country. I know with increasing certainty that my country will be poor for the rest of my life. I also expect AI to be as available as computers: there are the "have", and there are the "don't have", which is almost always a lifetime condition.
This is why Anthropic wants these chinese AI models banned as they are in the lead in the AI race to zero and they know that there is no modal moat.
So don't tell Dario.
Every industry-wide scale technological revolution has happened because government funded a technology and then opened it up to the masses. Just look at your iPhone: GPS, the internet, AI voice assistants, touchscreens, microprocessors, lithium-ion batteries, etc all came from gov't research (I'm counting Bell Labs' gov't mandated monopoly + research funding as gov't)
Economist Mariana Mazzucato wrote a great book about this called The Entrepreneurial State: Debunking Public vs. Private Sector Myths
I really don't think China cares about that. Chinese government's governance logic is making everything so cheap that everyone can get and use it. They did it with EVs and other things. Now they are doing it with the AI.
You are basically paying out the nose for a few seconds of VRAM residence if you are giving significant money for cache reads.
The very nature of autoregressive language modeling is that every single output token produced "reads" the cache.
So in principle the price floor for a cache hit is the flat cost of 1 output token.
Now in reality it has to be more than that because you are occupying VRAM with the cache that forces out other users. But it can still be really cheap.
And using up gpus for that cache is a pretty big opportunity cost. I highly doubt it's done in vram. That would be insane for the one hour caches.
So its memory + the time it takes to unload/load into vram + the extra cost per output token
Is it a scam? Idk
They aren't aiming companies but users which many have no common sense and grant these agentic AI access to everything.
All the restrictions the US imposed to CH, will be reverted back and it will be even worse, because now the data is not reaching the US gov ( we all know they have access to US big techs data ) but CH.
I really hope this goes viral and breaks Nvidia/OpenAI.