Weights are on Huggingface FWIW.
<a href="https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro&#x2F;tree&#x2F;main" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro&#x2F;tree&#x2F;main</a>

there&#x27;s no pinging. Someone&#x27;s gotta email dang

My submission here <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47885014">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47885014</a> done at the same time was to the weights.dang, probably the two should be merged and that be the link

For those who didn&#x27;t check the page yet, it just links to the API docs being updated with the upcoming models, not the actual model release.

<a href="https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Flash-Base" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Flash-Base</a>
<a href="https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro-Base" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro-Base</a>And we got new base models, wonderful, truly wonderful

Weights available here: <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro</a>

There appears a flight of dragons without heads. Good fortune.That&#x27;s literally what the I Ching calls &quot;good fortune.&quot;Competition, when no single dragon monopolizes the sky, brings fortune for all.

The thing is, it doesnt need to beat 4.7. it just needs to do somewhat well against it.This is free... as in you can download it, run it on your systems and finetune it to be the way you want it to be.

It is more than good enough and has effectively caught up with Opus 4.6 and GPT 5.4 according to the benchmarks.It&#x27;s about 2 months behind GPT 5.5 and Opus 4.7.As long as it is cheap to run for the hosting providers and it is frontier level, it is a very competitive model and impressive against the others. I give it 2 years maximum for consumer hardware to run models that are 500B - 800B quantized on their machines.It should be obvious now why Anthropic really doesn&#x27;t want you to run local models on your machine.

Tbh I was more productive with 4.6 than ever before and if AI progress locks in permanently at 4.6 tier, I’d be pretty happy

How does it compare to Opus 4.7? I&#x27;ve been immersed in 4.7 all week participating in the Anthropic Opus 4.7 hackathon and it&#x27;s pretty impressive even if it&#x27;s ravenous from a token perspective compared to 4.6

Their Chinese announcement says that, based on internal employee testing, it is not as good as Opus 4.6 Thinking, but is slightly better than Opus 4.6 without Thinking enabled.

Apparently glm5.1 and qwen coder latest is as good as opus 4.6 on benchmarks. So I tried both seriously for a week (glm Pro using CC) and qwen using qwen companion. Thought I could save $80 a month. Unfortunately after 2 days I had switched back to Max. The speed (slower on both although qwen is much faster) and errors (stupid layout mistakes, inserting 2 footers then refusing to remove one, not seeing obvious problems in screenshots &amp; major f-ups of functionality), not being able to view URLs properly, etc. I&#x27;ll give deepseek a go but I suspect it will be similar. The model is only half the story. Also been testing gpt5.4 with codex and it is very almost as good as CC... better on long running tasks running in background. Not keen on ChatGPT codex &#x27;personality&#x27; so will stick to CC for the most part.

Is it honestly better than Opus 4.6 or just benchmaxxed? Have you done any coding with an agent harness using it?If its coding abilities are better than Claude Code with Opus 4.6 then I will definitely be switching to this model.

Anthropic fans would claim God itself is behind Opus by 3-6 months and then willingly be abused by Boris and one of his gaslighting tweets.LMAO

&gt; (better than Opus 4.6)There we go again :) It seems we have a release each day claiming that. What&#x27;s weird is that even deepseek doesn&#x27;t claim it&#x27;s better than opus w&#x2F; thinking. No idea why you&#x27;d say that but anyway.Dsv3 was a good model. Not benchmaxxed at all, it was pretty stable where it was. Did well on tasks that were ood for benchmarks, even if it was behind SotA.This seems to be similar. Behind SotA, but not by much, and at a much lower price. The big one is being served (by ds themselves now, more providers will come and we&#x27;ll see the median price) at 1.74$ in &#x2F; 3.48$ out &#x2F; 0.14$ cache. Really cheap for what it offers.The small one is at 0.14$ in &#x2F; 0.28$ out &#x2F; 0.028$ cache, which is pretty much &quot;too cheap to matter&quot;. This will be what people can run realistically &quot;at home&quot;, and should be a contender for things like haiku&#x2F;gemini-flash, if it can deliver at those levels.

This resonates with me a lot.I do some stuff with gemini flash and Aider, but mostly because I want to avoid locking myself into a walled garden of models, UIs and company

What do you run these on? I&#x27;ve gotten comfortable with Claude but if folks are getting Opus performance for cheaper I&#x27;ll switch.

Agree with your assessment, I think after models reached around Opus 4.5 level, its been almost indistinguishable for most tasks. Intelligence has been commoditized, what&#x27;s important now is the workflows, prompting, and context management. And that is unique to each model.

I don’t know what people are doing but Minimax produced 16 bugreports which of 15 was false positives (literally a mistake).In contrast ChatGPT 5.3 and also Opus has a 90% rate at least on this same project. (Embedded)All other tests were the same. What are you doing with these models?

Which model&#x27;s best depends on how you use it. There&#x27;s a huge difference in behaviour between Claude and GPT and other models which makes some poor substitutes for others in certain use cases. I think the GPT models are a bad substitute for Claude ones for tasks such as pair-programming (where you want to see the CoT and have immediate responses) and writing code that you actually want to read and edit yourself, as opposed to just letting GPT run in the background to produce working code that you won&#x27;t inspect. Yes, GPT 5.4 is cheap and brilliant but very black-box and often very slow IME. GPT-5.4 still seems to behave the same as 5.1, which includes problems like: doesn&#x27;t show useful thoughts, can think for half an hour, says &quot;Preparing the patch now&quot; then thinks for another 20 min, gives no impression of what it&#x27;s doing, reads microscopic parts of source files and misses context, will do anything to pass the tests including patching libraries...

actually this is not the reason - the harness is significantly better.
There is no comparable harness to Claude Code with skills, etc.Opencode was getting there, but it seems the founders lost interest. Pi could be it, but its very focused on OpenClaw. Even Codex cli doesnt have all of it.which harness works well with Deepseek v4 ?

eh idk. until yesterday opus was the one that got spatial reasoning right (had to do some head pose stuff, neither glm 5.1 nor codex 5.3 could &quot;get&quot; it) and codex 5.3 was my champion at making UX work.So while I agree mixed model is the way to go, opus is still my workhorse.

Is Opus nerfed somehow in Copilot? Ive tried it numerous times, it has never reallt woved me. They seem to have awfully small context windows, but still. Its mostly their reasoning which has been offCodex is just so much better, or the genera GPT models.

I found Opus 4.7 to be actually worse than Opus 4.6 for my use caseSubstantially worse at following instructions and overoptimized for maximizing token usage

I don&#x27;t think we need to compare models to Opus anymore. Opus users don&#x27;t care about other models, as they&#x27;re convinced Opus will be better forever. And non-Opus users don&#x27;t want the expense, lock-in or limits.As a non-Opus user, I&#x27;ll continue to use the cheapest fastest models that get my job done, which (for me anyway) is still MiniMax M2.5. I occasionally try a newer, more expensive model, and I get the same results. I have a feeling we might all be getting swindled by the whole AI industry with benchmarks that just make it look like everything&#x27;s improving.

For the curious, I did some napkin math on their posted benchmarks and it racks up 20.1 percentage point difference across the 20 metrics where both were scored, for an average improvement of about 2% (non-pp). I really can&#x27;t decide if that&#x27;s mind blowing or boring?Claude4.6 was almost 10pp better at at answering questions from long contexts (&quot;corpuses&quot; in CorpusQA and &quot;multiround conversations&quot; in MRCR), while DSv4 was a staggering 14pp better at one math challenge (IMOAnswerBench) and 12pp better at basic Q&amp;A (SimpleQA-Verified).

No, the Deepseek V4 paper itself says that DS-V4-Pro-Max is close to Opus 4.5 in their staff evaluations, not better than 4.6:&gt; In our internal evaluation, DeepSeek-V4-Pro-Max outperforms Claude Sonnet 4.5 and approaches the level of Opus 4.5.

<a href="https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro&#x2F;blob&#x2F;main&#x2F;DeepSeek_V4.pdf" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro&#x2F;blob&#x2F;main...</a>Model was released and it&#x27;s amazing. Frontier level (better than Opus 4.6) at a fraction of the cost.

MErge? <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47885014">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47885014</a>

It is funny seeing people ping pong between Anthropic and ChatGPT, with similar rhetoric in both directions.At this point I would just pick the one who&#x27;s &quot;ethics&quot; and user experience you prefer. The difference in performance between these releases has had no impact on the meaningful work one can do with them, unless perhaps they are on the fringes in some domain.Personally I am trying out the open models cloud hosted, since I am not interested in being rug pulled by the big two providers. They have come a long way, and for all the work I actually trust to an LLM they seem to be sufficient.

I didn&#x27;t express this well but my interest isn&#x27;t &quot;who is in the top spot&quot;, and is more _why and _how various labs get the results they do. This is also magnified by the fact that I&#x27;m not only interested in hosted providers of inference but local models as well. What&#x27;s your take on the best model to run for coding on 24GB of VRAM locally after the last few weeks of releases? Which harness do you prefer? What quants do you think are best? To use your sports metaphor it&#x27;s more than following the national leagues but also following college and even high school leagues as well. And the real interest isn&#x27;t even who&#x27;s doing well but WHY, at each level.

The players barely ever change. People don&#x27;t have problems following sports, you shouldn&#x27;t struggle so much with this once you accept top spot changes.

This is only good advice if you don’t have the need to understand what’s happening on the edge of the frontier. If you do, then you’ll lose on compounding the knowledge from staying engaged with the major developments.

This one’s been particularly hard to sit out because the executive and managerial class are absolutely mainlining this stuff and pushing it hard on the rest of the organization, and so whether or not I want to keep up, I need to, because my job is to actually make stuff work and this stuff is a borderline existential risk to the quality of the systems I’m responsible for and rely on.

Don&#x27;t keep up. Much like with news, you&#x27;ll know when you need to know, because someone else will tell you first.

For coding Opus 4.5 in q3 2025 was still the best model I&#x27;ve used.Since then it&#x27;s just been a cycle of the old model being progressively lobotomised and a &quot;new&quot; one coming out that if you&#x27;re lucky might be as good as the OG Opus 4.5 for a couple of weeks.Subjective but as far as I can tell no progress in almost a year, which is a lifetime in 2022-25 LLM timelines

I don&#x27;t trust the benchmarks either, so I maintained a set of benchmarks myself. I&#x27;m mostly interested in local models, and for the past 2 years they have steadily gotten better.Can&#x27;t argue with subjective experience, but if there were some tasks that you thought LLMs can&#x27;t do two years ago, maybe try again today. You might be surprised.

It honestly has all kinda felt like more of the same ever since maybe GPT4?New model comes out, has some nice benchmarks, but the subjective experience of actually using it stays the same. Nothing&#x27;s really blown my mind since.Feels like the field has stagnated to a point where only the enthusiasts care.

<a href="https:&#x2F;&#x2F;commoncog.com&#x2F;how-to-make-sense-of-ai&#x2F;" rel="nofollow">https:&#x2F;&#x2F;commoncog.com&#x2F;how-to-make-sense-of-ai&#x2F;</a>

I’m very satisfied with being three months behind everything in AI.  That’s a level that’s useful, the overhyped nonsense gets found out before I need to care, and it’s easy enough to keep up with.

I’m deeply interested and invested in the field but I could really use a support group for people burnt out from trying to keep up with everything. I feel like we’ve already long since passed the point where we need AI to help us keep up with advancements in AI.

Excited that the long awaited v4 is finally out. But feel sad that it&#x27;s not multimodal native.

Insert obligatory &quot;this is the way&quot; Mando scene. Indeed!

doesn&#x27;t it get tiring after a while? using the same (perceived) gotcha, over and over again, for three years now?no one is ever going to release their training data because it contains every copyrighted work in existence. everyone, even the hecking-wholesome safety-first Anthropic, is using copyrighted data without permission to train their models. there you go.

Aww yes, let me push a couple petabytes to my git repo for everyone to download...

Weights are the source, training data is the compiler.

They are exactly open source. The training data is the internet. Don&#x27;t say it&#x27;s on the internet. It IS the internet.The training scripts are in Megatron and vLLM.

Where&#x27;s the training data and training scripts since you are calling this open source?Edit: it seems &quot;open source&quot; was edited out of the parent comment.

Their audience is people who build stuff, techs audience is enterprise CEOs and politicians, and anyone else happy to hype up all the questionably timed releases and warnings of danger, white collar irrelevence, or promises of utopian paradise right before a funding round.

If those docs were written by Deepseek, it’s also a pretty positive review of the model.

now that we can use AI to write the docs , test the docs , proof read the docs , it really isn&#x27;t that much of a feat right?

There&#x27;s something heartwarming about the developer docs being released before the flashy press release.

Yeah if I could use Gemini with pi.dev that would be my choice. But Gemini CLI is just so, so bad.

The sheer number of bugs and lack of meaningful improvements in Google products is a clear counterargument to the AI bull thesisIf AI was so good at coding, why can’t it actually make a usable Gemini&#x2F;AI Studio app?

I gave up on Gemini 3.1 Pro in VSCode after 2 hours. They fully refunded me.

Most of these tests are one-prompt in nature. I&#x27;ve also noticed issues with the PDF reader in Gemini which was very frustrating, although it is significantly better now than it was even two weeks ago. On the contrary, now GPT-5 seems to be giving me issues.In my experience, Gemini is the most insightful model for hard problems (particularly math problems that I work on).

You know, with a bit of prompting, you can instruct Gemini to output the state of the conversation into a prompt that you can enter in a new chat and continue where you left off. But now with a fresh context window.

Funny how Gemini is theoretically the best -- but in practice all the bugs in the interface mean I don&#x27;t want to use it anymore. The worst is it forgets context (and lies about it), but it&#x27;s very unreliable at reading pdfs (and lies about it). There&#x27;s also no branch, so once the context is lost&#x2F;polluted, you have to start projects over and build up the context from scratch again.

MMLU-Pro:Gemini-3.1-Pro at 91.0Opus-4.6 at 89.1GPT-5.4, Kimi2.6, and DS-V4-Pro tied at 87.5Pretty impressive

SOTA MRCR (or would&#x27;ve been a few hours earlier... beaten by 5.5), I&#x27;ve long thought of this as the most important non-agentic benchmark, so this is especially impressive. Beats Opus 4.7 here

I hope the update is an improvement. Losing 3.2 would be a real loss, it&#x27;s excellent.

More like he wants to ban accelerator chip sales to China, which may be about “national security” or self preservation against a different model for AI development which also happens to be an existential threat to Anthropic. Maybe those alternatives are actually one and the same to him.

Annecotal, but I saw a tweet from someone who interviewed at Anthropic, and was explicity rejected because of cultural mismatch because they were not against open weight models.It&#x27;s hard not to see Anthropic&#x27;s messaging of &quot;this tech that we&#x27;re pushing on you is going to take your job and maybe kill you&quot; as being about anything other than regulatory capture, with the goal of the government shutting down competitors.I think OpenAI and Anthropic are both really in a tough spot - spending so much on what is becoming a commodity product for which neither seems positioned to be low cost producer. Maybe a bit like the UK-France channel tunnel project where the product itself is a success but a bloodbath for those who invested to build it.

&gt; this is why Anthropic wants to ban open weight modelsDo you have a source?

The paper is here: [0]Was expecting that the release would be this month [1], since everyone forgot about it and not reading the papers they were releasing and 7 days later here we have it.One of the key points of this model to look at is the optimization that DeepSeek made with the residual design of the neural network architecture of the LLM, which is manifold-constrained hyper-connections (mHC) which is from this paper [2], which makes this possible to efficiently train it, especially with its hybrid attention mechanism designed for this.There was not that much discussion around it some months ago here [3] about it but again this is a recommended read of the paper.I wouldn&#x27;t trust the benchmarks directly, but would wait for others to try it for themselves to see if it matches the performance of frontier models.Either way, this is why Anthropic wants to ban open weight models and I cannot wait for the quantized versions to release momentarily.[0] <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro&#x2F;blob&#x2F;main&#x2F;DeepSeek_V4.pdf" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro&#x2F;blob&#x2F;main...</a>[1] <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47793880">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47793880</a>[2] <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2512.24880" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2512.24880</a>[3] <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=46452172">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=46452172</a>

more like weekly or almost daily, gpt 5.5 was literally 12 hours ago

At this point &#x27;frontier model release&#x27; is a monthly cadence, Kimi 2.6 Claude 4.6 GPT 5.5, the interesting question is which evals will still be meaningful in 6 months.

Any visualised benchmark&#x2F;scoreboard for comparison between latest models? DeepSeek v4 and GPT-5.5 seems to be ground breaking.

OpenCode? Pi?If one finds it difficult to set up OpenCode to use whatever providers they want, I won&#x27;t call them &#x27;dev&#x27;.The only real friction (if the model is actually as good as SOTA) is to convince your employer to pay for it. But again if it really provides the same value at a fraction of the cost, it&#x27;ll eventually cease to be an issue.

They have instructions right on their page on how to use claude code with it.

You can literally run it from Claude code. Easily too

It&#x27;s difficult because even if the underlying model is very good, not having a pre-built harness like Claude Code makes it very un-sticky for most devs. Even at equal quality, the friction (or at least perceived friction) is higher than the mainstream models.

History doesn&#x27;t always repeat itself.But if it does, then in the following week we&#x27;ll see DeepSeek4 floods every AI-related online space. Thousands of posts swearing how it&#x27;s better than the latest models OpenAI&#x2F;Anthropic&#x2F;Google have but only costs pennies.Then a few weeks later it&#x27;ll be forgotten by most.

Do you have the links for those? Very interested

I&#x27;ve been calling that the &quot;streaming experts&quot; trick, the key idea is to take advantage of Mixture of Expert models where only a subset of the weights are used for each round of calculations, then load those weights from SSD into RAM for each round.As I understand it if DeepSeek v4 Pro is a 1.6T, 49B active that means you&#x27;d need just 49B in memory, so ~100GB at 16 bit or ~50GB at 8bit quantized.v4 Flash is 284B, 13B active so might even fit in &lt;32GB.

These are more like experiments than a polished release as of yet. And the reduction in throughput is high compared to having the weights in RAM at all times, since you&#x27;re bottlenecked by the SSD which even at its fastest is much slower than RAM.

Weren&#x27;t there some frameworks recently released to allow Macs to stream weights from fast SSDs and thus fit way more parameters than what would normally fit in RAM?I have never tried one yet but I am considering trying that for a medium sized model.

&quot;2 minutes ago&quot; <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;unsloth&#x2F;DeepSeek-V4-Pro" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;unsloth&#x2F;DeepSeek-V4-Pro</a>

Unsloth often turn them around within a few hours, they might have gone to bed already though!Keep an eye on <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;unsloth&#x2F;models" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;unsloth&#x2F;models</a>Update ten minutes later: <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;unsloth&#x2F;DeepSeek-V4-Pro" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;unsloth&#x2F;DeepSeek-V4-Pro</a> just appeared but doesn&#x27;t have files in yet, so they are clearly awake and pushing updates.

How long does it usually take for folks to make smaller distills of these models? I really want to see how this will do when brought down to a size that will run on a Macbook.

Output is what the compute is used for above all else; costs more hardware time basically than prompt processing (input) which is a lot faster

input tokens are processed at 10-50 times the speed of output tokens since you can process then in batches and not one at a time like output tokens

Price is appealing to me. I have been using gemini 3 flash mainly for chat. I may give it a try.input: $0.14&#x2F;$0.28 (whereas gemini $0.5&#x2F;$3)Does anyone know why output prices have such a big gap?

I&#x27;m going to blow my bandwidth allowance again this month, aren&#x27;t I.

The Flash version is 284B A13B in mixed FP8 &#x2F; FP4 and the full native precision weights total approximately 154 GB. KV cache is said to take 10% as much space as V3. This looks very accessible for people running &quot;large&quot; local models. It&#x27;s a nice follow up to the Gemma 4 and Qwen3.5 small local models.

Better link:<a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47885014">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47885014</a><a href="https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro</a>

The Flash one should - it&#x27;s 160GB on Hugging Face: <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Flash&#x2F;tree&#x2F;main" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Flash&#x2F;tree&#x2F;ma...</a>

Which version fits in a Mac Studio M3 Ultra 512 GB?

Please don&#x27;t slander the most open AI company in the world. Even more open than some non-profit labs from universities. DeepSeek is famous for publishing everything. They might take a bit to publish source code but it&#x27;s almost always there. And their papers are extremely pro-social to help the broader open AI community. This is why they struggle getting funded because investors hate openness. And in China they struggle against the political and hiring power of the big tech companies.Just this week they published a serious foundational library for LLMs <a href="https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;TileKernels" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;TileKernels</a>Others worth mentioning:<a href="https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;DeepGEMM" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;DeepGEMM</a> a competitive foundational library<a href="https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;Engram" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;Engram</a><a href="https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;DeepSeek-V3" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;DeepSeek-V3</a><a href="https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;DeepSeek-R1" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;DeepSeek-R1</a><a href="https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;DeepSeek-OCR-2" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;deepseek-ai&#x2F;DeepSeek-OCR-2</a>They have 33 repos and counting: <a href="https:&#x2F;&#x2F;github.com&#x2F;orgs&#x2F;deepseek-ai&#x2F;repositories?type=all" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;orgs&#x2F;deepseek-ai&#x2F;repositories?type=all</a>And DeepSeek often has very cool new approaches to AI copied by the rest. Many others copied their tech. And some of those have 10x or 100x the GPU training budget and that&#x27;s their moat to stay competitive.The models from Chinese Big Tech and some of the small ones are open weights only. (and allegedly benchmaxxed) (see <a href="https:&#x2F;&#x2F;xcancel.com&#x2F;N8Programs&#x2F;status&#x2F;2044408755790508113" rel="nofollow">https:&#x2F;&#x2F;xcancel.com&#x2F;N8Programs&#x2F;status&#x2F;2044408755790508113</a>). Not the same.

Weights are the source, training data is the compiler

&gt; Internet comments say that open sourcing is a national strategy, a loss maker subsidized by the government. On the contrary, it is a commercial strategy and the best strategy available in this industry.This sounds whole lot like potatoh potahto. I think the former argument is very much the correct one: China can undercut everyone and win, even at a loss. Happened with solar panels, steel, evs, sea food - it&#x27;s a well tested strategy and it works really well despite the many flavors it comes in.That being said a job well done for the wrong reasons is still a job well done so we should very much welcome these contributions, and maybe it&#x27;s good to upset western big tech a bit so it&#x27;s remains competitive.

if you want to understand why labs open source their models: <a href="http:&#x2F;&#x2F;try.works&#x2F;why-chinese-ai-labs-went-open-and-will-remain-open" rel="nofollow">http:&#x2F;&#x2F;try.works&#x2F;why-chinese-ai-labs-went-open-and-will-rema...</a>

It’s not remotely hypothetical you’d have to be living under a rock to believe that. And the fusion with a one-party state government that doesn’t tolerate huge swathes of thoughtspace being freely discussed is completely streamlined, not mediated by any guardrails or accountability.This “no harm to me” meme about a foreign totalitarian government (with plenty of incentive to run influence ops on foreigners) hoovering your data is just so mind-bogglingly naive.

As someone with Tibetan friends and as someone from India, Chinese ulterior motives are way more clear.

I, personally, have never been asked for an asshole scan, but I&#x27;m interested in providing one if you can point me to a company that&#x27;s offering.

American companies want a scan of your asshole for the privilege of paying to access their models, and unapologetically admit to storing, analyzing, training on, and freely giving your data to any authorities if requested. Chinese ulteriority is hypothetical, American is blatant.

I don’t care about whatever “ulterior motives” they might haveMy country’s per capita income is $2500 a year. We can’t pay perpetual rent to OAI&#x2F;Anthropic

Lol EU pats you on the headIts sad to see how you have regulated yourselves into a position where Mistral is your only claim.

It&#x27;s a little sad that tech now comes down to geopolitics, but if you&#x27;re not in the USA then what is the difference? I&#x27;m Danish, would I rather give my data to China or to a country which recently threatened the kingdom I live in with military invasion? Ideally I&#x27;d give them to Mistral, but in reality we&#x27;re probably going to continue building multi-model tools to make sure we share our data with everyone equally.

Do they also open-source censoring filter rules? Like, you can&#x27;t ask what happened at Tiananmen Square in 1989.

I think their motives are pretty transparent, as are china’s, as ever, you have to pick the lesser of two evils.

&gt; I know if the potential ulterior motives.And you think the US tech giants don&#x27;t have any ulterior motives?!

How are the &quot;ulterior motives&quot; of Chinese companies any worse than &quot;ulterior motives&quot; of US companies or European ones?

Truly open source coming from China.  This is heartwarming.  I know if the potential ulterior motives.

That&#x27;s right, need a lot of GPUs + memory, anybody experimented with Mac Studio M3 Ultra for this?

They have released mixed fp8&#x2F;fp4 for efficiency. It&#x27;s still hundreds of gigabytes, though. Give up on local for these.

Its on OR - but currently not available on their anthropic endpoint. OR if you read this, pls enable it there! I am using kimi-2.6 with Claude Code, works well, but Deepseek V4 gives an error:`<a href="https:&#x2F;&#x2F;openrouter.ai&#x2F;api&#x2F;messages" rel="nofollow">https:&#x2F;&#x2F;openrouter.ai&#x2F;api&#x2F;messages</a> with model=deepseek&#x2F;deepseek-v4-pro, OR returns
 an error because their Anthropic-compat translator doesn&#x27;t cover V4 yet. The Claude CLI dutifully surfaces that error as &quot;model...does not exist&quot;

<a href="https:&#x2F;&#x2F;openrouter.ai&#x2F;deepseek&#x2F;deepseek-v4-pro" rel="nofollow">https:&#x2F;&#x2F;openrouter.ai&#x2F;deepseek&#x2F;deepseek-v4-pro</a><a href="https:&#x2F;&#x2F;openrouter.ai&#x2F;deepseek&#x2F;deepseek-v4-flash" rel="nofollow">https:&#x2F;&#x2F;openrouter.ai&#x2F;deepseek&#x2F;deepseek-v4-flash</a>

Try interacting with it through the website, it will give an error and some explanation on the issue.  I had to relax my guardrail settings.

Getting &#x27;Api Error&#x27; here :(
Every other model is working fine.

Yup, can&#x27;t really be used in production atm.

The Pro model is giving 429 Overload errors

Already on Openrouter. Pro version is $1.74&#x2F;m&#x2F;input, $3.48&#x2F;m&#x2F;output, while flash $0.14&#x2F;m&#x2F;input, 0.28&#x2F;m&#x2F;output.

The same way you fit a bucket wheel excavator in your garage

The low end could be something like an eBay-sourced server with a truckload of DDR3 ram doing all-cpu inference - secondhand server models with a terabyte of ram can be had for about 1.5K. The TPS will be absolute garbage and it will sound like a jet engine, but it will nominally run.The flash version here is 284B A13B, so it might perform OK with a fairly small amount of VRAM for the active params and all regular ram for the other params, but I’d have to see benchmarks. If it turns out that works alright, an eBay server plus a 3090 might be the bang-for-buck champ for about $2.5K (assuming you’re starting from zero).

Flash is less than 160 GB. No need to quantize to fit in 2x 96 GB. Not sure how much context fits in 30 GB, but it should be a good amount.

For flash? 4 bit quant, 2x 96GB gpu (fast and expensive) or 1x 96GB gpu + 128GB ram (still expensive but probably usable, if you’re patient).A mac with 256 GB memory would run it but be very slow, and so would be a 256GB ram + cheapo GPU desktop, unless you leave it running overnight.The big model? Forget it, not this decade. You can theoretically load from SSD but waiting for the reply will be a religious experience.Realistically the biggest models you can run on local-as-in-worth-buying-as-a-person hardware are between 120B and 200B, depending on how far you’re willing to go on quantization. Even this is fairly expensive, and that’s before RAM went to the moon.

A loaded macbook pro can get you to the frontier from 24 months ago at ~10-40tok&#x2F;s, which is plenty fast enough for regular chatting.

Run on an old HEDT platform with a lot of parallel attached storage (probably PCIe 4) and fetch weights from SSD.  You&#x27;d ultimately be limited by the latency of these per-layer fetches, since MoE weights are small. You could reduce the latencies further by buying cheap Optane memory on the second-hand market.

Look at GB&#x2F;s.Strix halo has 256 GB&#x2F;s bandwidth for $2500.
The Flash model has 13 GB activations.256 &#x2F; 13 = 19.6 tokens per secondExcept you cannot fit it into the maximum RAM of 128 GB Strix Halo supports. So move on.Another option is Threadripper. That&#x27;s 8 memory channels. Using older DDR4-3200 you get roughly 200 GB&#x2F;s. For $2000.200 &#x2F; 13 = 15.4 tokens per secondBut, a chunk of per-token weights is actually always the same and not MoE, so you would offload that to a GPU and get a decent speedup. Say 25 tokens per second total.Then likely some expensive Mac. No idea.Eventually you arrive at a mining rig chassis with a beefy board and multiple GPUs. That has the benefit of pipelining. You run part of the model on one GPU and move on, so another batch can start on the first one. Low (say 30-100) tps individually, but a lot more in parallel. Best get it with other people.

How can you reasonably try to get near frontier (even at all tps) on hardware you own? Maybe under 5k in cost?

yeah. look at these 4 feathers (?) on his bum too.

The pro pelican is a work of art! It goes dimensions that no other LLM has gone before.

No way. The Pro pelican is fatter, has a customized front fork, and the sun is shining! He’s definitely living the best life.

The Flash one is pretty impressive. Might be my favorite so far in the pelican-riding-a-bicycle series

I really like the pro version. The pelican is so cute.

This is an excellent comment. Thanks for this - I&#x27;ve only ever thought about whether the frame is the right shape, I never thought about how different illustrations might map to different bicycle categories.

The Pedersen looks like someone failed the &quot;draw a bicycle&quot; test and decided to adjust the universe.

Being a bicycle geometry nerd I always look at the bicycle first.Let me tell you how much the Pro one sucks... It looks like failed Pedersen[1]. The rear wheel intersects with the bottom bracket, so it wouldn&#x27;t even roll. Or rather, this bike couldn&#x27;t exist.The flash one looks surprisingly correct with some wild fork offset and the slackest of seat tubes. It&#x27;s got some lowrider[2] aspirations with the small wheels, but with longer, Rivendellish[3], chainstays. The seat post has different angle than the seat tube, so good luck lowering that.[1] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Pedersen_bicycle" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Pedersen_bicycle</a>[2] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Lowrider_bicycle" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Lowrider_bicycle</a>[3] <a href="https:&#x2F;&#x2F;www.rivbike.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.rivbike.com&#x2F;</a>

<a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47879092#47880421">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47879092#47880421</a>

They&#x27;re stressed pelicans from Hangzhou.

DeepSeek pelicans are the angriest pelicans I’ve seen so far.

It was a bigger deal in the Gemini 3.1 launch: <a href="https:&#x2F;&#x2F;x.com&#x2F;JeffDean&#x2F;status&#x2F;2024525132266688757" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;JeffDean&#x2F;status&#x2F;2024525132266688757</a>

I think the pelican on a bike is known widely enough that of seizes to be useful as a benchmark. There is even a pelican briefly appearing in the promo video of GPT-5, if I&#x27;m not mistaken <a href="https:&#x2F;&#x2F;openai.com&#x2F;gpt-5&#x2F;" rel="nofollow">https:&#x2F;&#x2F;openai.com&#x2F;gpt-5&#x2F;</a>. So the companies are apparently aware of it.

&gt;Generate an SVG of a pelican riding a bicycleat the top of the linked pages.

What was your prompt for the image? Apologies if this should be obvious.

This should be the bottom comment on the pelican comment on every model release post.

This should not be the top comment on every model release post. It&#x27;s getting tiring.

To me this is the perfect proof that1) LLM is not AGI. Because surely if AGI it would imply that pro would do better than flash?2) and because of the above, Pelican example is most likely already being benchmaxxed.

Is it then Deepseek hosted by Deepseek?How much does the drawing change if you ask it again?

I tried that for the GPT-5 launch - a self-improving loop that renders the SVG, looks at it and tries again - and the results were surprisingly disappointing.I should try it again with the more recent models.

This is just a random thought, but have you tried doing an &#x27;agentic&#x27; pelican?As in have the model consider its generated SVG, and gradually refine it, using its knowledge of the relative positions and proportions of the shapes generated, and have it spin for a while, and hopefully the end result will be better than just oneshotting it.Or maybe going even one step further - most modern models have tool use and image recognition capabilities - what if you have it generate an SVG (or parts&#x2F;layers of it, as per the model&#x27;s discretion) and feed it back to itself via image recognition, and then improve on the result.I think it&#x27;d be interesting to see, as for a lot of models, their oneshot capability in coding is not necessarily corellated with their in-harness ability, the latter which really matters.

I like the pelican I got out of deepseek-v4-flash more than the one I got from deepseek-v4-pro.<a href="https:&#x2F;&#x2F;simonwillison.net&#x2F;2026&#x2F;Apr&#x2F;24&#x2F;deepseek-v4&#x2F;" rel="nofollow">https:&#x2F;&#x2F;simonwillison.net&#x2F;2026&#x2F;Apr&#x2F;24&#x2F;deepseek-v4&#x2F;</a>Both generated using OpenRouter.For comparison, here&#x27;s what I got from DeepSeek 3.2 back in December: <a href="https:&#x2F;&#x2F;simonwillison.net&#x2F;2025&#x2F;Dec&#x2F;1&#x2F;deepseek-v32&#x2F;" rel="nofollow">https:&#x2F;&#x2F;simonwillison.net&#x2F;2025&#x2F;Dec&#x2F;1&#x2F;deepseek-v32&#x2F;</a>And DeepSeek 3.1 in August: <a href="https:&#x2F;&#x2F;simonwillison.net&#x2F;2025&#x2F;Aug&#x2F;22&#x2F;deepseek-31&#x2F;" rel="nofollow">https:&#x2F;&#x2F;simonwillison.net&#x2F;2025&#x2F;Aug&#x2F;22&#x2F;deepseek-31&#x2F;</a>And DeepSeek v3-0324 in March last year: <a href="https:&#x2F;&#x2F;simonwillison.net&#x2F;2025&#x2F;Mar&#x2F;24&#x2F;deepseek&#x2F;" rel="nofollow">https:&#x2F;&#x2F;simonwillison.net&#x2F;2025&#x2F;Mar&#x2F;24&#x2F;deepseek&#x2F;</a>

It is tricky to build good infrastructure for prompt caching.

I wonder why there aren&#x27;t more open weights model with support for prompt caching on OpenRouter.

For comparison on openrouter DeepSeek v4 Flash is slightly cheaper than Gemma 4 31b, more expensive than Gemma 4 26b, but it does support prompt caching, which means for some applications it will be the cheapest. Excited to see how it compares with Gemma 4.

Probably usable for chat sessions, unusably slow for agentic coding.

Unironically curious about the performance of this model on unified VRAM machines.

Oh well, I should have bought 2x 512GB RAM MacStudios, not just one :(

You can, but does it work well? I assume CC has all kinds of Claude specific prompts in it, wouldn&#x27;t you be better with a harness designed to be model agnostic like pi.dev or OpenCode?

<a href="https:&#x2F;&#x2F;opencode.ai&#x2F;" rel="nofollow">https:&#x2F;&#x2F;opencode.ai&#x2F;</a>

<a href="https:&#x2F;&#x2F;pi.dev&#x2F;" rel="nofollow">https:&#x2F;&#x2F;pi.dev&#x2F;</a>

You can use CC with other models, you aren’t forced to use Claude model.

What&#x27;s the current best framework to have a &#x27;claude code&#x27; like experience with Deepseek (or in general, an open-source model), if I wanted to play?

If they offer something close to Z.ai:s coding plan during Christmas, I’ll take it!

Anyone tried with make web UI with it? How good is it? For me opus is only worth because of it.

A few hours after GPT5.5 is wild. Can’t wait to try it.

We will be hosting it soon at getlilac.com!

Right now Im much more worried about sending data to the US and A.. At least theres a less chanse it will be missused against -me-

The chances of my bank account getting hacked due to the PLA backdoor in Deepseek is higher than the CIA backdoor in OpenAI.

&gt; I am uncomfortable about sending user data which may contain PII to their servers in ChinaAs a European I feel deeply uncomfortable about sending data to US companies where I know for sure that the government has access to it.I also feel uncomfortable sending it to China.If you&#x27;d asked me ten years ago which one made me more uncomfortable. China.But now I&#x27;m not so sure, in fact I&#x27;m starting to lean towards the US as being the major risk.

Kimi is indeed somewhat cheap for frontier-level intelligence, but still is $4-5 per mm tokens. Deep Seek is at least an order of magnitude cheaper.

&gt; For context, for an agent we&#x27;re working on, we&#x27;re using 5-mini, which is $2&#x2F;1m tokens. This is $0.30&#x2F;1m tokens. And it&#x27;s Opus 4.6 level - this can&#x27;t be real.It&#x27;s doesn&#x27;t seem all that out there compared to the other Chinese model price&#x2F;performance? Kimi2.6 is cheaper even than this, and is pretty close in performance

Interesting, thanks. I&#x27;ll keep an eye out.

Since it&#x27;s open weights it&#x27;ll be available on AWS Bedrock soon(ish), likely at a higher price than the official API but still coming in under those GPT-5-mini prices.

OpenAI released a privacy filter model <a href="https:&#x2F;&#x2F;openai.com&#x2F;index&#x2F;introducing-openai-privacy-filter" rel="nofollow">https:&#x2F;&#x2F;openai.com&#x2F;index&#x2F;introducing-openai-privacy-filter</a> . It is small enough to run locally, I would pass my request through it first before sending to any apis.

This is shockingly cheap for a near frontier model. This is insane.For context, for an agent we&#x27;re working on, we&#x27;re using 5-mini, which is $2&#x2F;1m tokens. This is $0.30&#x2F;1m tokens. And it&#x27;s Opus 4.6 level - this can&#x27;t be real.I am uncomfortable about sending user data which may contain PII to their servers in China so I won&#x27;t be using this as appealing as it sounds. I need this to come to a US-hosted environment at an equivalent price.Hosting this on my own + renting GPUs is much more expensive than DeepSeek&#x27;s quoted price, so not an option.

Using it with opencode sometimes it generates commands like:<pre><code> bash({&quot;command&quot;:&quot;gh pr create --title &quot;Improve Calendar module docs and clean up idiomatic Elixir&quot; --body &quot;$(cat &lt;&lt;&#x27;EOF&#x27;
 Problem
 The Calendar modu...
</code></pre>
like generating output, but not actually running the bash command so not creating the PR ultimately. I wonder if it&#x27;s a model thing, or an opencode thing.

Google stated a while back, that with tpus they are able to sell at cost &#x2F; with profit.Aka: everyone who uses Nvidia isn&#x27;t selling at cost, because Nvidia is so expensive.

My thoughts exactly. I also believe that subscription services are profitable, and the talk about subsidies is just a way to extract higher profit margins from the API prices businesses pay.

In six month deepseek won&#x27;t be sota anymore und usage will be wayyyy down.

This price is high even because of the current shortage of inference cards available to DeepSeek; they claimed in their press release that once the Ascend 950 computing cards are launched in the second half of the year, the price of the Pro version will drop significantly

As this is a new arch with tons of optimisations, it&#x27;ll take some time for inference engines to support it properly, and we&#x27;ll see more 3rd party providers offer it. Once that settles we&#x27;ll have a median price for an optimised 1.6T model, and can &quot;guesstimate&quot; from there what the big labs can reasonably serve for the same price. But yeah, it&#x27;s been said for a while that big labs are ok on API costs. The only unknown is if subscriptions were profitable or not. They&#x27;ve all been reducing the limits lately it seems.

These models are open and there are tons of western providers offering it at comparable rates.

not soooo much though. It&#x27;s heavily subsidized for residential consumption, but industrial power rates are almost comparable to the US (depends on the state you go to etc).

Point taken but there isnt any western providers there yet. Power is cheaper in china.

Amazingly, the current depreciation overestimates the retained value of GPUs.In 2023, the depreciation schedule for H100s was 2 years, but they are still oversubscribed and generating signficant income.Coreweve has upped their depreciation for GPUs to 6 years(!) now, which seems more realistic.<a href="https:&#x2F;&#x2F;www.silicondata.com&#x2F;blog&#x2F;h100-rental-price-over-time" rel="nofollow">https:&#x2F;&#x2F;www.silicondata.com&#x2F;blog&#x2F;h100-rental-price-over-time</a>

They are profitable to opex costs, but not capex costs with the current depreciation schedules, though those are now edging higher than expected.

Before the AI bubble that will burst any time now, there was the AI winter that would magically arrive before the models got good enough to rival humans.

It is possible to question the sustainability of the AI buildout and not have a dogmatic position on AI development.There are still major unanswered questions here. For instance, all of the incremental data capacity build out is going to businesses that have totally unknown LT unit economics and that today are burning obscene amounts of cash.

The people who doubted the sustainability of dot com era bubbles were correct even though the tech was actually transformational. Personally I expect roughly the same outcome.

Insert always has been meme.But seriously, it just stems from the fact some people want AI to go away. If you set your conclusion first, you can very easily derive any premise. AI must go away -&gt; AI must be a bad business -&gt; AI must be losing money.

When China does it it&#x27;s communism. When companies in the west get massive tax cuts, rebates, incentives and subsidies, that&#x27;s just supporting the captains of industry.

&gt; I’d like somebody to explain to me how the endless comments of &quot;bleeding edge labs are subsidizing the inference at an insane rate&quot; make sense in light of a humongous model like v4 pro being $4 per 1M. I’d bet even the subscriptions are profitable, much less the API prices.One answer - Chinese Communist Party. They are being subsidized by the state.

And they actually say the prices will be &quot;significantly&quot; lower in second semester when Huawei 650 chips comes in.

you know, if you don&#x27;t have to pay insane salary for your top engineers, and don&#x27;t have to pay billions for internet shills to control the narrative, then all of the labs will be insane profitable.

I mean, not one &quot;bleeding edge&quot; lab has stated they are profitable. They don&#x27;t publish financials aside from revenue. And in Anthropic&#x27;s case, they fuck with pricing every week. Clearly something is wrong here.

They’ve also announced Pro price will further drop 2H26 once they have more HUAWEI chips.

I&#x27;m pretty sure OpenAI and Anthropic are overpricing their token billed API usage mainly as an incentive to commit to get their subscriptions instead.

I was thinking the same. How can it be than other providers can offer third-party open source models with roughly the similar quality like this, Kimi K2.6 or GLM 5.1 for 10 times less the price? How can it be that GPT 5.5 is suddenly twice the price as GPT 5.4 while being faster? I don&#x27;t believe that it&#x27;s a bigger, more expensive model to run, it&#x27;s just they&#x27;re starting to raise up the prices because they can and their product is good (which is honest as long as they&#x27;re transparent with it). Honestly the movement about subscription costing the company 20 times more than we&#x27;re paying is just a PR movement to justify the price hike.

Research and training costs have to be amortized from somewhere; and labs are always training. I&#x27;m definitely keen for the financials when the two files for IPO though, it would be interesting to see; although I&#x27;m sure it won&#x27;t be broken down much.

API prices may be profitable. Subscriptions may still be subsidized for power users. Free tiers almost certainly are. And frontier labs may be subsidizing overall business growth, training, product features, and peak capacity, even if a normal metered API call is profitable on marginal inference.

Yeah, subscriptions used to be extraordinarily generous. I miss those days, but the reinvigoration of open weight models is super exciting.I&#x27;m still playing with the new Qwen3.6 35B and impressed, now DeepSeek v4 drops; with both base and instruction-tuned weights? There goes my weekend :P

The claim that APIs are subsidized is very common.eg:Token prices are significantly subsidized and anyone that does any serious work with AI can tell you this.<a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47684887">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47684887</a>(the claims don&#x27;t make any sense, but they are widely held)

I haven&#x27;t seen anyone claiming that API prices are subsidized.At some point (from the very beginning till ~2025Q4) Claude Code&#x27;s usage limit was so generous that you can get roughly $10~20 (API-price-equivalent) worth of usage out of a $20&#x2F;mo Pro plan each day (2 * 5h window) - and for good reason, because LLM agentic coding is extremely token-heavy, people simply wouldn&#x27;t return to Claude Code for the second time if provided usage wasn&#x27;t generous or every prompt costs you $1. And then Codex started trying to poach Claude Code users by offering even greater limits and constantly resetting everyone&#x27;s limit in recent months. The API price would have to be 30x operating cost to make this not a subsidy. That would be an extraordinary claim.

It&#x27;s the decades of performance doesn&#x27;t matter SV&#x2F;web culture. I&#x27;d be surprised if over 1% of OpenAI&#x2F;Anthropic staff know how any non-toy computer system works.

It&#x27;s because investors in OpenAI&#x2F;Anthropic want to get their money back in 10 months, not in 10 years.

They got loans to buy inference hardware on the promise of potential AGI, or at least something approaching ASI, all leading to stupid amounts of profit for those investors.We therefore cannot just look at inference costs directly, training is part of the pitch. Without the promises of continuous improvement and chasing the elusive AGI, money for investments for inference evaporates.

Prices are not just hard cost of inference. Training costs are not equal. Chinese labs have cheaper access to large data centers. I also suspect they operate far more efficiently than orgs like openAI.

This is nonsense. The real reason is because the US companies are scamming the public, as per usual.

Because you are comparing China to the US.In China you need to appease state goals. In the US you need to appease investor goals.China will keep funding them regardless of their income, because the goal is (ostensibly) a state AGI&#x2F;ASI. In the US, the goal is an ROI which may or may not come with AGI&#x2F;ASI.They are different economies with different goals. We can look at past Chinese national projects and see that they are fine with burning $50 to get [social goal] that&#x27;s worth $5.

&gt; pricing &quot;Pro&quot; $3.48 &#x2F; 1M output tokens vs $4.40I’d like somebody to explain to me how the endless comments of &quot;bleeding edge labs are subsidizing the inference at an insane rate&quot; make sense in light of a humongous model like v4 pro being $4 per 1M. I’d bet even the subscriptions are profitable, much less the API prices.edit: $1.74&#x2F;M input
$3.48&#x2F;M output on OpenRouter

<a href="https:&#x2F;&#x2F;api-docs.deepseek.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;api-docs.deepseek.com&#x2F;</a><a href="https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro&#x2F;blob&#x2F;main&#x2F;DeepSeek_V4.pdf" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;deepseek-ai&#x2F;DeepSeek-V4-Pro&#x2F;blob&#x2F;main...</a>