I've been using DeepSeek V4 a lot in the last week and I am very happy with it. If you have a really gnarly bug, you might need a SOTA model like Opus. For most things it is very very good, and costs significantly less (even without the discount).
I've been using it as part of a complex DOS game decompilation project[0]. I'm working on refactoring the software rendering pipeline so that we can add GPU rendering. The hardest part of this so far is converting the 90's polygon rendering from screen to world space.
It spun its wheels a few times doing a large mostly mechanical change. After resetting and improving my prompts it was able to get through it. I'm using Matt Pocock's skills[1] for this work, which has been quite nice.
This is clearly a well-timed loss-leading strategic market share grab! Anthropic have blown a lot of user trust in the last couple of months..
But, overall, the current AI pricing is completely unsustainable, across all AI companies, except via the exponential growth they are relying on. Dylan Patel did the most insightful analysis of this I've come across.. https://youtu.be/mDG_Hx3BSUE?si=nyJu4adwYCH1igbJ
Really feel like the current versions are for sure "good enough". Thats not how market capture is gonna function though and they are gonna keep pushing because the only moat is to stay ahead, so the problems gonna stay strange. at some point more compute isn't a reasonable answer, and optimization is, and my feeling is we are well past that point from a product perspective, but ipos etc etc
The only moat is the us trying to buy all the compute hardware in the world for the next two years. Then China, amd, etc are just making their own chips.
So I think the current generation of models are arguably all about the same in terms of capability. However, the requirement for exponential growth I mentioned is all about the economics.
AI companies are trying to ride a growth wave where the income curve lags the expense curve by 1-2 years, and at the same time investing 10x their historical income on next year's projected demand.
Everyone is selling their API calls at a loss, because to capture the investment required to scale the business up and the costs down, you need to grow your market now (in relative and absolute terms). And history shows, that in big tech you often have winner-takes-all situations, or, at least a couple of big firms will dominate, and the others will die. That's where market share becomes a key strategic goal.
But to secure that, they also need to be building next year's compute now. And if their anticipated compute needs are 10x this year, they've got a serious funding problem, one that can only be filled by capital with an appropriate risk appetite. You can only get this high-risk capital when the potential payoff is even more enormous, or, when it's a smaller bite of a much bigger pie. Hence, MS putting into OpenAI and so on. But the investment needs are getting so big we are starting to see some pullback from more conservative sources, but also record deals from others.
Now say an AI company does get the capital they need to grow. Well, they've still got a very serious supply problem. RAM, GPUs, water, electricity etc. Hence why there's a lot of deals and cross-investment going on - everyone is trying to secure resources and lower their overall risk exposure while keeping a foot in every possible door, so they can switch alliances whenever it's expedient, and because collaboration also helps the overall market to grow.
This all explains to me why the industry _needs_ the hype. These companies can't exist without it, because the money they need to sink in, in order to even be around in 18 months, far outstrips all reasonable financial practices. So it's capitalism on steroids or nothing. If you believe the AI story, then to that extent, it's rational.
But note that nowhere in this scenario does it suggest the actual consumers will be getting a consistent product at a consistent price!!!
Qwen 3.6 supports reasonable agentic programming. People are vibe coding with it. It's really not that far off. If you truly cannot make a model that was SOTA 6-12 months ago work for you today for free I don't want to know what your needs are.
a lack of existential threat in the form of pay-seeking and remediation from the people you stole training materials from that allows for an intrinsically different pace of operation than the Western competition
I’m not happy with their privacy policy [1]. I’m unfamiliar with the phrase “Parties with Other Legal Rights”. Given the well-documented struggles of Anthropic and others to provide enough compute, I wonder if “Parties with Other Legal Rights” constitutes part of the advantage here.
Just run a local model or run deepseek from another provider with a policy you like. The models are open weight and widely available. Still cheaper than chatgpt and anything else through 3rd parties
this is the pitch - it's open source, run it yourself. But >99% of people will not have the hardware needed to run these models at a high enough quality to be close to SOTA. So they will run the open-source models on CCP systems for a good price.
They need to build data centers and lots of them everywhere, preferably powered with renewable energy. Let the tokens flow like water. The models are finally getting to the point where the LLM just knows what you’re asking for and gives it to you.
Free lunch? More like "free data". The fools who give their life data and most intimate Intellectual property over to the AI companies for free, yes that's a free lunch that won't be subsidized for much longer when the cost on them which has been unsustainable (their data being harvested for non-training purposes) come stop catch up with them.
Sincerely,
- I see you AI companies harvesting our data giving us discounted subscriptions so we can not realize we are paying you to take our own data!
there will be free lunch till they admit to themselves that there is no moat. Acquring customers at huge costs is a fools errand when models are mostly indisguishable.
Anthropic is learning that lesson now. Doesnt help that their ceo goes around antognozing everyone by claiming jobs are over and annoying boris does like 500 podcasts per week repeating "coding is solved"
I can't figure out how there's both too little supply (so a dramatic need for more data centers) but also too little demand (so labs subsidize inference).
There isn't too little demand. There is massive demand and many competing companies trying to capture that demand, so they are attempting to make better offers than their competition. Hence subsidy.
- Every competitor is planning for the demand to be much higher in a few years than it is now, and aiming to capture as much of that as they can, which starts by getting companies hooked on their models now
- The data center capacity will get used no matter who captures the most demand
I find it hard to understand why nobody in this thread considers that the current pricing might still be below cost. The discount was supposed to end on May 5, and then shortly after that they extended it to May 31. They clearly made a judgment call there, rather than treating it as a desperate loss-leader.
If you have actually used DeepSeek, you would notice that the cache-hit rate is extremely high, and the cache invalidation window is much longer than every other provider's. That suggests DeepSeek is simply much better at utilizing its infrastructure than other vendors.
I am also highly skeptical that the average user's input is worth more than the API cost of processing it. Do people really think DeepSeek researchers enjoy panning for gold in a river of boilerplate and half-baked code?
Unless you're very careful, it's trivial to have my secrets to be sent to the LLM. If it reads your .env just to see the variable names, the secrets have been sent to the servers. Now - they probably don't care about you and your secrets - but it makes me more uncomfortable that they have them.
This is true of anthropic or openai - but for some reason I think the us govt or anyone else will have a harder time getting to my data from them than the CCP will any chinese company.
> If it reads your .env just to see the variable names, the secrets have been sent to the servers.
But these are development secrets. They're of no use to anyone.
Production secrets probably live in another repo, or at least are stored encrypted. In my case both. And I'm managing the least sensitive stuff you can imagine: a pre-school website, a non-profit event website, personal blog...
> for some reason I think the us govt or anyone else will have a harder time getting to my data
US companies are required by law to hand over your data if given a warrant by USG. They don't need a warrant if they have a subpoena for less invasive data, or a FISA request. They can also ask without any justification, and see if the company will cough it up anyway (they often do). Any AI company with government contracts will want to give up data quicker so as not to threaten deals worth hundreds of millions.
> but for some reason I think the us govt or anyone else will have a harder time getting to my data from them than the CCP will any chinese company.
US tech companies voluntarily give their data to the US government. Don't you remember PRISM? You think they stopped doing that?
> Internal NSA presentation slides included in the various media disclosures show that the NSA could unilaterally access data and perform "extensive, in-depth surveillance on live communications and stored information" with examples including email, video and voice chat, videos, photos, voice-over-IP chats (such as Skype), file transfers, and social networking details.[2] Snowden summarized that "in general, the reality is this: if an NSA, FBI, CIA, DIA, etc. analyst has access to query raw SIGINT [signals intelligence] databases, they can enter and get results for anything they want."[13]
>I think the us govt or anyone else will have a harder time getting to my data than the CCP will any chinese company.
Why? You dont think that 5 eyes cyber peeps use every advantage they can get? And on the way out leave a dusting of evidence pointing at the russkies or chinese?
Why would two companies burning 100s of billions of dollars and are not profitable be safe keepers of your data when there is a huge market for all of that in the us and the us has really weak protections for those things so the companies can sell it to defense agencies?
Thing is, either way your data is getting hoovered up. If not today then eventually. It's just a matter of where. If you work in an industry where nation states might want to do you irreparable harm then yea don't let your data leave the country.
It's unlikely that you're special enough that someone will genuinely look through the massive amount of data produced by this system in order to target You Specifically. If you are that special you can just use another provider.
From this line of reasoning, my guess is that the huge discount is not so much intended to sell the data collection system as much as it is intended to sell the model. If you had to wring a geopolitical consequence from this, it would be that the US labs producing models would be impacted by a vastly less expensive competitor.
Are you actually planning on travelling out of the country right now? It's probably not a good idea even if you don't use Chinese products, which by the way you definitely do.
The people that travel out of the country are typically not the same ones aligned with the current administration. The vast majority of the MAGA base are more likely to not have a passport, while a large portion have probably never left their state.
I don't think anyone will ever be confused by the "only country to use the bomb" in this context. Your pedantry is not something to not feel bad about as it does nothing constructive to the conversation
only different if you don't consider "the bomb" the way it is colloquially used. until today, i would have never considered saying "the bomb" would not have been understood as a nuclear weapon whether that was an a-bomb or an h-bomb.
I mean I can't believe I have to say this explicitly but it should be assumed that any data you send to China can and will be used against our interest by the CCP...
Kicking myself when my little vibe coded widget to notify me when socks go on sale that does not and never has functioned properly is wielded as a mighty scepter to topple western hegemony
Pi (pi.dev) is fine. I'm using it with DS v4 right now. It's not close to Claude code but I think that's the point.
By the way OpenRouter version is very slow for some reason. DeepSeek platform is faster (and cheaper with the discount) if you don't mind passing the credit card number / email to this company.
I've been using DeepSeek V4 a lot in the last week and I am very happy with it. If you have a really gnarly bug, you might need a SOTA model like Opus. For most things it is very very good, and costs significantly less (even without the discount).
I've been using it as part of a complex DOS game decompilation project[0]. I'm working on refactoring the software rendering pipeline so that we can add GPU rendering. The hardest part of this so far is converting the 90's polygon rendering from screen to world space.
It spun its wheels a few times doing a large mostly mechanical change. After resetting and improving my prompts it was able to get through it. I'm using Matt Pocock's skills[1] for this work, which has been quite nice.
[0]: https://github.com/FatalDecomp/ROLLER
[1]: https://github.com/mattpocock/skills
What agentic harness do you use deepseek with?
I'm using Pi: https://github.com/badlogic/pi-mono/tree/main/packages/codin...
A few days ago we were hearing about how the "free lunch is over", now we're seeing discounts and increased usage limits.
This is clearly a well-timed loss-leading strategic market share grab! Anthropic have blown a lot of user trust in the last couple of months..
But, overall, the current AI pricing is completely unsustainable, across all AI companies, except via the exponential growth they are relying on. Dylan Patel did the most insightful analysis of this I've come across.. https://youtu.be/mDG_Hx3BSUE?si=nyJu4adwYCH1igbJ
Really feel like the current versions are for sure "good enough". Thats not how market capture is gonna function though and they are gonna keep pushing because the only moat is to stay ahead, so the problems gonna stay strange. at some point more compute isn't a reasonable answer, and optimization is, and my feeling is we are well past that point from a product perspective, but ipos etc etc
The only moat is the us trying to buy all the compute hardware in the world for the next two years. Then China, amd, etc are just making their own chips.
So I think the current generation of models are arguably all about the same in terms of capability. However, the requirement for exponential growth I mentioned is all about the economics.
AI companies are trying to ride a growth wave where the income curve lags the expense curve by 1-2 years, and at the same time investing 10x their historical income on next year's projected demand.
Everyone is selling their API calls at a loss, because to capture the investment required to scale the business up and the costs down, you need to grow your market now (in relative and absolute terms). And history shows, that in big tech you often have winner-takes-all situations, or, at least a couple of big firms will dominate, and the others will die. That's where market share becomes a key strategic goal.
But to secure that, they also need to be building next year's compute now. And if their anticipated compute needs are 10x this year, they've got a serious funding problem, one that can only be filled by capital with an appropriate risk appetite. You can only get this high-risk capital when the potential payoff is even more enormous, or, when it's a smaller bite of a much bigger pie. Hence, MS putting into OpenAI and so on. But the investment needs are getting so big we are starting to see some pullback from more conservative sources, but also record deals from others.
Now say an AI company does get the capital they need to grow. Well, they've still got a very serious supply problem. RAM, GPUs, water, electricity etc. Hence why there's a lot of deals and cross-investment going on - everyone is trying to secure resources and lower their overall risk exposure while keeping a foot in every possible door, so they can switch alliances whenever it's expedient, and because collaboration also helps the overall market to grow.
This all explains to me why the industry _needs_ the hype. These companies can't exist without it, because the money they need to sink in, in order to even be around in 18 months, far outstrips all reasonable financial practices. So it's capitalism on steroids or nothing. If you believe the AI story, then to that extent, it's rational.
But note that nowhere in this scenario does it suggest the actual consumers will be getting a consistent product at a consistent price!!!
We're subsidized by the Chinese government!
https://www.reuters.com/world/asia-pacific/deepseek-nears-45...
Cool go download qwen 3.6 and run it on a single GPU and you can avoid paying into a subsidized model
why are we pretending these are equivalents?
yes, single gpu open models exist. Now show me the one that can keep up with a SOTA api model on more than short code block evals.
Qwen 3.6 supports reasonable agentic programming. People are vibe coding with it. It's really not that far off. If you truly cannot make a model that was SOTA 6-12 months ago work for you today for free I don't want to know what your needs are.
People don't understand that deep seek is running a plausibly sustainable business. Like how qwen/Alibaba is.
Every AI vendor is trying to steal marketshare. For now the competition is good!
I'm guessing there was a pullback in usage as the free lunch started ending. So we get some more subsidized usage.
* from Chinese labs
What advantage do you think they have?
Operating in a jurisdiction where US companies can't sue them.
a lack of existential threat in the form of pay-seeking and remediation from the people you stole training materials from that allows for an intrinsically different pace of operation than the Western competition
I’m not happy with their privacy policy [1]. I’m unfamiliar with the phrase “Parties with Other Legal Rights”. Given the well-documented struggles of Anthropic and others to provide enough compute, I wonder if “Parties with Other Legal Rights” constitutes part of the advantage here.
[1]: https://cdn.deepseek.com/policies/en-US/deepseek-privacy-pol...
Just run a local model or run deepseek from another provider with a policy you like. The models are open weight and widely available. Still cheaper than chatgpt and anything else through 3rd parties
this is the pitch - it's open source, run it yourself. But >99% of people will not have the hardware needed to run these models at a high enough quality to be close to SOTA. So they will run the open-source models on CCP systems for a good price.
What I mean is you can use providers who also host deepseek models for pennies without touching deepseek itself.
I’m only seeing 3x the cost of DeepSeek for other providers on Open Router. Is there a better place to look?
I haven't really had this issue but deepinfra claims to have us servers and looks pretty cheap to me.
A sane government policy that invests heavily on innovative businesses.
They need to build data centers and lots of them everywhere, preferably powered with renewable energy. Let the tokens flow like water. The models are finally getting to the point where the LLM just knows what you’re asking for and gives it to you.
Free lunch? More like "free data". The fools who give their life data and most intimate Intellectual property over to the AI companies for free, yes that's a free lunch that won't be subsidized for much longer when the cost on them which has been unsustainable (their data being harvested for non-training purposes) come stop catch up with them.
Sincerely, - I see you AI companies harvesting our data giving us discounted subscriptions so we can not realize we are paying you to take our own data!
there will be free lunch till they admit to themselves that there is no moat. Acquring customers at huge costs is a fools errand when models are mostly indisguishable.
Anthropic is learning that lesson now. Doesnt help that their ceo goes around antognozing everyone by claiming jobs are over and annoying boris does like 500 podcasts per week repeating "coding is solved"
I can't figure out how there's both too little supply (so a dramatic need for more data centers) but also too little demand (so labs subsidize inference).
There isn't too little demand. There is massive demand and many competing companies trying to capture that demand, so they are attempting to make better offers than their competition. Hence subsidy.
That, and:
- Every competitor is planning for the demand to be much higher in a few years than it is now, and aiming to capture as much of that as they can, which starts by getting companies hooked on their models now
- The data center capacity will get used no matter who captures the most demand
Per 1M tokens (input cache hit / input cache miss / output)
v4-pro (75% off): $0.003625 / $0.435 / $0.87
v4-pro (regular): $0.0145 / $1.74 / $3.48
v4-flash: $0.0028 / $0.14 / $0.28
that is damn cheap.
You are the product. The book is called "So long, and thanks for all the secrets"
What if I told you.. this was no different to every US company
You are the product whenever you are sending your data to an LLM not controlled by you.
Nothing specific to Deepseek.
Generous of you to think I'm doing top secret coding and not just another cat website
I find it hard to understand why nobody in this thread considers that the current pricing might still be below cost. The discount was supposed to end on May 5, and then shortly after that they extended it to May 31. They clearly made a judgment call there, rather than treating it as a desperate loss-leader.
If you have actually used DeepSeek, you would notice that the cache-hit rate is extremely high, and the cache invalidation window is much longer than every other provider's. That suggests DeepSeek is simply much better at utilizing its infrastructure than other vendors.
I am also highly skeptical that the average user's input is worth more than the API cost of processing it. Do people really think DeepSeek researchers enjoy panning for gold in a river of boilerplate and half-baked code?
Is anyone concerned about these services and China’s National Intelligence Law?
No because China can only do so much to me as someone who doesn't live there and never will.
It's the same reason why I prefer vpns that are owned by countries outside my own.
Unless you're very careful, it's trivial to have my secrets to be sent to the LLM. If it reads your .env just to see the variable names, the secrets have been sent to the servers. Now - they probably don't care about you and your secrets - but it makes me more uncomfortable that they have them.
This is true of anthropic or openai - but for some reason I think the us govt or anyone else will have a harder time getting to my data from them than the CCP will any chinese company.
> If it reads your .env just to see the variable names, the secrets have been sent to the servers.
But these are development secrets. They're of no use to anyone.
Production secrets probably live in another repo, or at least are stored encrypted. In my case both. And I'm managing the least sensitive stuff you can imagine: a pre-school website, a non-profit event website, personal blog...
> for some reason I think the us govt or anyone else will have a harder time getting to my data
US companies are required by law to hand over your data if given a warrant by USG. They don't need a warrant if they have a subpoena for less invasive data, or a FISA request. They can also ask without any justification, and see if the company will cough it up anyway (they often do). Any AI company with government contracts will want to give up data quicker so as not to threaten deals worth hundreds of millions.
> but for some reason I think the us govt or anyone else will have a harder time getting to my data from them than the CCP will any chinese company.
US tech companies voluntarily give their data to the US government. Don't you remember PRISM? You think they stopped doing that?
> Internal NSA presentation slides included in the various media disclosures show that the NSA could unilaterally access data and perform "extensive, in-depth surveillance on live communications and stored information" with examples including email, video and voice chat, videos, photos, voice-over-IP chats (such as Skype), file transfers, and social networking details.[2] Snowden summarized that "in general, the reality is this: if an NSA, FBI, CIA, DIA, etc. analyst has access to query raw SIGINT [signals intelligence] databases, they can enter and get results for anything they want."[13]
>I think the us govt or anyone else will have a harder time getting to my data than the CCP will any chinese company.
Why? You dont think that 5 eyes cyber peeps use every advantage they can get? And on the way out leave a dusting of evidence pointing at the russkies or chinese?
Why would two companies burning 100s of billions of dollars and are not profitable be safe keepers of your data when there is a huge market for all of that in the us and the us has really weak protections for those things so the companies can sell it to defense agencies?
Thing is, either way your data is getting hoovered up. If not today then eventually. It's just a matter of where. If you work in an industry where nation states might want to do you irreparable harm then yea don't let your data leave the country.
It's unlikely that you're special enough that someone will genuinely look through the massive amount of data produced by this system in order to target You Specifically. If you are that special you can just use another provider.
From this line of reasoning, my guess is that the huge discount is not so much intended to sell the data collection system as much as it is intended to sell the model. If you had to wring a geopolitical consequence from this, it would be that the US labs producing models would be impacted by a vastly less expensive competitor.
Not for my purposes tbh. Enjoy my shitty javascript, Xi.
Should be TypeScript
yes. imagine getting denied at the border or something because of data you shared with deep seek,WeChat or any other china centric service
Are you actually planning on travelling out of the country right now? It's probably not a good idea even if you don't use Chinese products, which by the way you definitely do.
The people that travel out of the country are typically not the same ones aligned with the current administration. The vast majority of the MAGA base are more likely to not have a passport, while a large portion have probably never left their state.
Definitely would select the frowny face if that happened.
Might as well answer yes to the "are you a subversive" question
the US does that to you too, for not liking your opinions about particular parties or intelligence aparati.
I think martin_henk is fully aware of that and it's why of all the examples of how a government can use your data, he picked this one...
No I'm more concerned with OpenAI and Anthropic AI models being used as a tool to murder brown people in the middle east for our "greatest ally".
More worried about the Epstein regime
Eh I’m using it for stuff where there is nothing proprietary or identifiable.
Eh, I'd be more concerned about the Three-Letters and the One country that dropped an A-bomb.
>the One country that dropped an A-bomb.
i'd like to point out that the soviet RDS-3 was an airdropped A-bomb.
I get that you mean 'in anger', but I don't feel that bad being a pedant against a propagandist statement that's also pedantically wrong.
I don't think anyone will ever be confused by the "only country to use the bomb" in this context. Your pedantry is not something to not feel bad about as it does nothing constructive to the conversation
> the One country that dropped an A-bomb
> "only country to use the bomb"
I was confused. Two very different statements but I assume they refer to the US who dropped two A-bombs (Hiroshima and Nagasaki) in anger.
only different if you don't consider "the bomb" the way it is colloquially used. until today, i would have never considered saying "the bomb" would not have been understood as a nuclear weapon whether that was an a-bomb or an h-bomb.
I mean I can't believe I have to say this explicitly but it should be assumed that any data you send to China can and will be used against our interest by the CCP...
Yes! The only saving grace is that they have so many secrets, mine are not so important.
Kicking myself when my little vibe coded widget to notify me when socks go on sale that does not and never has functioned properly is wielded as a mighty scepter to topple western hegemony
Same with the USA. Difference is China is not bombing brown-skinned people every so often.
"our"?
What coding agent(ideally CLI) have people found works well with this?
Occasionally I go and try different agents with openrouter models, but nothing seems to really get close to the proprietary ones like claude-code.
Pi (pi.dev) is fine. I'm using it with DS v4 right now. It's not close to Claude code but I think that's the point.
By the way OpenRouter version is very slow for some reason. DeepSeek platform is faster (and cheaper with the discount) if you don't mind passing the credit card number / email to this company.
Have any regular Opus users taken V4 for a spin? What’s your take?
Anecdatally, out of all the popular LLMs I’ve only found Gemini to be any use for entry-level Ford Power Stroke Diesel mechanics and diagnostics. :)
lmao i can pay them to steal my ideas and code
[dead]