DeepSeek-V4-Flash Update (api-docs.deepseek.com)
631 points by dnhkng 14 hours ago
NitpickLawyer 14 hours ago
This is more exciting than k3, IMO. Dsv4 models are extremely cheap to serve. Improving their capabilities has lots of downstream effects, as it becomes "good enough" for more and more tasks.
DS was serving the pro version at extremely low prices for a long time, and they've had integrations with opencode & other providers, so they likely gathered a lot of data from real developers doing real tasks (on openrouter they were labeled as such). Now they can use those live scenarios to further post-train their models and improve them further.
Can't wait to see if distilling k3 into dsv4 brings additional improvements. Anyway, having fast cheap models getting better is great for the community. Especially since these don't "go away" on a provider's whim. Whatever capabilities they get, can be used "forever" going forward. And, at least flash can be ran "at home" with <10k in hardware, which isn't really possible / feasible with glm/k3 larger models.
dotancohen 3 hours ago
> Whatever capabilities they get, can be used "forever" going forward.
sigzero 2 hours ago
I upvoted you just for the Dune reference.
idiotsecant an hour ago
chorizo 5 hours ago
Starting to wonder if the free big pickle model on opencode has been DSV4F0731 for the past few months. It’s been incredibly fast and good.
kevincox 4 hours ago
At least in the past it was GLM-4.6. IDK if it is ever changed.
chorizo 4 hours ago
dnhkng 14 hours ago
Totally! This with DwarfStar delivers usable local AI (I hope!)
dannyw 11 hours ago
Usable local AI has been here for a while, esp on say a 5090.
You can’t treat Qwen3.6 like its fable, but if you prompt precisely and specifically it’s a great executor.
I actually found it refreshing to use more of my brain for once, and actually have to think deeper about what I’m trying to do, and how to build it.
genxy 3 hours ago
Parent is referring to https://github.com/antirez/ds4
Tepix 10 hours ago
What are your goalposts? Depending on your requirements, there have been many moments of usable local AI. More recent ones were gpt-oss 120b and Qwen 3.6 27b.
KaseyKim 11 hours ago
hope that deepseek become better
ilaksh 10 hours ago
Have you tried the one that was just released?
f311a 13 hours ago
I've been driving flash model for 90% of my tasks. It's better than pro (for unknown reasons), very cheap and fast.
I try to keep changes under 1000 lines and drive architectural decisions myself, barely notice any difference compared to frontier models. The rest 10% is to spot bugs, security problems and to investigate better architecture, which flash can also do pretty well, I just cross check it.
Faster iterations are way better for me, I hate waiting for 5-10 minutes on small changes. I tried to use recent versions of Kimi and GLM, but they use too much thinking for no reason and are pretty slow because of it. I also often feed a lot of data to it, without worrying about hitting the limits: dependencies (to find bottlenecks in them), logs, performance dumps and so on.
Also, it will never complain about security guards, I've been using it to reverse engineer binaries.
embedding-shape 12 hours ago
> Also, it will never complain about security guards, I've been using it to reverse engineer binaries.
Maybe I'm using too weak language in my prompts, but none of the OpenAI models I've used via codex has refused to reverse engineer binaries, is it supposed to? I'm sitting right now reverse-engineering a 3rd party firmware together with Codex and haven't hit a single guardrail. Meanwhile, I see people complaining about it rejecting non-security related prompts, are things so individual on the platforms right now or what's going on?
markasoftware 12 hours ago
Have you completed the identity verification? It's much more lenient once you have
embedding-shape 12 hours ago
flexagoon 12 hours ago
realusername 12 hours ago
I got an account warning on OpenAI (waved after I complained) just because I was asking it how to root some >10 years old Android device.
dotancohen 3 hours ago
mark_l_watson 9 hours ago
This is my experience also: DeepSeek v4 flash is good enough for most of my work and I like the fast response times. I buy tokens mostly from FireWorks.ai in the US, but I also prepaid for a large chunk of tokens directoy with DeepSeek.
I use OpenCode mostly (uses fewer tokens than Claude Code) and I am looking forward to the release of DeepSeek’s own coding harness.
regularfry 12 hours ago
It's good but (at least on openrouter) it's got an annoyingly tight output token limit. So if it does get stuck in a reasoning pit, it won't work its way out of it in time.
It's replaced the Kimi models for me though.
u8080 12 hours ago
Try to use DS platform directly - cheaper and better than openrouter, no subscription
dghlsakjg 6 hours ago
lionkor 12 hours ago
I use deepseek for a lot of my personal day-to-day agent needs, and I will simply put this here and let this speak for itself, last 30 days:
- Cost: $4.55USD
- API requests: 3,467
- Tokens: 323,183,886
And as an engineer who leads a small team, I have very high standards for quality, and these carry across to my personal projects where I use deepseek. It has not disappointed at all for coding or review tasks. For everything else, use another model.
troglodytetrain 2 hours ago
DeepSeek is amazing, they are, from a cost/benefit literally an order of magnitude or more better than the 'SOTA' models, and yet no one really talks about them.
I'm using them for my micro-saas, and they have made my niche economically profitable where as SOTA models are only slightly better for massively increased expense. Its truly impressive.
Word of advice to anyone, not all your use of LLM tech needs to be code/dev work related.
We are entering 'Web 4.0 era' or whatever you want to call it. Massive transformations of nearly every single business will and are being developed as the cost of intelligence as a commodity is falling through the floor...
a20eac1d 10 hours ago
Can you give more info on how you use/prompt those LLMs for code review and what kind of prompts you use?
I've had worse experiences doing it because the quality of answer has been quite bad, and I'm wondering if my methods are the reason.
lionkor 9 hours ago
Yes, gladly! I have not yet open-sourced my skills etc., but I can give some insight and share a couple here.
Review is a skill, as in, a SKILL.md with a folder full of references:
- SKILL.md: https://gist.github.com/lionkor/161525be858d1d75db4c13c0f093...
- references/output-contract.md: https://gist.github.com/lionkor/8c68e33becef7a21f8408c7dc119...
- references/review-lenses.md: https://gist.github.com/lionkor/0a8b080fe45306213efddf3ebb75...
- references/review-workflow.md: https://gist.github.com/lionkor/d2d374b133ceb7e3660bd530ee72...
- references/section-rules.md: https://gist.github.com/lionkor/8a9e503adc7fd3697410cf021f27...
I'm aware that almost all of this is prompt voodoo, and there's no guarantee for the review to find anything or everything, but making it a dedicated skill and thoroughly observing the output thinking, tool calls, and result, lets me adjust these over time and fill the weak spots with even more prompting.
I use this skill by simply telling the agent something like "Review the changes on the current branch against origin/main, take special care with backwards-incompatible changes to the public API" or something like that.
I use `pi` (pi.dev) with a subagents extension, so that I can ask the agent to invoke a subagent to do the review, on work that the agent did.
For models, I use the highest possible reasoning on whatever model I feel like makes sense, usually this is GPT-5.5 or deepseek flash/pro, depending on the confidentiality of the codebase, on the highest reasoning always (for reviews).
I've also had success with a review checklist, though it doesn't produce an easy to parse (for humans) output: https://gist.github.com/lionkor/054ac2cf241e0765eee2383f0dba...
This is why my review skill mandates a very strict output contract. I need the output to be very easy to parse, and the output contract I've specified there does that.
In general I let <whatever the latest model of OpenAI's ChatGPT is> author and review SKILL.md and similar large prompts, usually with a ruleset like this, which is a 1600 line research artifact from a long GPT 5.5 "Pro" research session on prompt engineering: https://gist.github.com/lionkor/71498794d0a7d72173fc58766f25...
Does the review catch all issues? Not at all. Does it catch, usually more than one, important issue, across large changesets? Absolutely, and that's the point! :)
Feel free to ask me any questions, I'm also happy to share more about my setup via email or add you or anyone else to my private repos with more of these.
rzerowan 3 hours ago
kekebo 5 hours ago
throwa356262 10 hours ago
What harness are you using to achieve that level of token caching?
lionkor 9 hours ago
I use pi, and, like the sibling comment, the caching ratio is fantastic. I work on C#, Rust, C, C++, shell scripting, and other areas.
k__ 9 hours ago
I'm using pi and my caching is ~99%.
brcmthrowaway 4 hours ago
OpenRouter?
lionkor 2 hours ago
No, platform.deepseek.com for me! The caching is super important, not sure how open router performs, I haven't tried it.
marcus_cemes an hour ago
innis226 11 hours ago
But doesn't it hallucinate a lot? Does that affect your workflow?
lionkor 9 hours ago
It hallucinates plenty, about the same as Codex models and all other LLMs! I review all code it writes, thoroughly, check the test coverage, write tests myself, have other models/chats cross-check the work with a review skill (which I've shared in another comment in this thread).
Usually unreviewed code only gets committed if I really don't care, like for one-off scripts, which I sandbox with github.com/lionkor/sbh or run as an unprivileged user.
maweaver 4 hours ago
You are using DeepSeek's services directly? Doesn't that end up sending at least snippets/chunks of code to a server where it is subject to Chinese government data access laws? Even if I was okay with that, my organization would never be. And even if they were, our partners/vendors/customers would not be. I think that's the sticking point for a lot of people.
lionkor 2 hours ago
Yes all of it goes straight to China! All of it is also open source or source available, or will be in the future. The stuff that isn't is trivial enough.
For any work with protected intellectual property, I use other providers, for the contractual guarantees, but I think it would be silly to think that OpenAI or Anthropic are not training on literally all data they get. How could you ever tell if they did? They can just claim the data was mislabelled, or ignore the accusations. If you have serious IP, use only local models.
anigbrowl 2 hours ago
So go with some other provider who hosts the models like OpenRouter if are worried about this.
kmarc 12 hours ago
Essentially I'm running everything on flash now inside pi. With the correct set of MCP servers, context reducer tooling and skills it can implement any task I throw at it. Some sessions take 30+ turns, but it's fast and cheap; all this in an hour, with ~$0.5 cost.
(TBH though, in my multi-subagent workflow I do use other, more expensive models for planning, reviewing, oracle-ing)
I haven't used our slow opus subscription for weeks.
(Also set up an OpenWebUi self-hosted chat that works from my phone, has some mcp and skills. fully replaced perplexity. Monthly cost ~$18 for hosting and subscriptions)
lionkor 12 hours ago
I want to second this, I use the same setup (pi + deepseek, with lots of custom tools for tracking TODOs, doing things with less tokens, etc, and with a SOTA model for very difficult tasks) and it's all I need it to be.
peperunas 12 hours ago
Do you have any recommendations of such extensions for pi?
sdesol 3 hours ago
This is self promotional but I am working on making pi extremely enterprise ready with:
https://github.com/gitsense/pi-brains/tree/staging
The README is being worked on but the three videos should give you a good sense of what it can do. Pi is also what makes what I will demo in
https://github.com/gitsense/chat/tree/update-readme
possible. Since Pi exposes so much, it is very easy to build advanced tooling around it to help easily grok hundreds of tool calls to help you understand what they agent knows and what it has tried.
kmarc 12 hours ago
Recommendation? No. Just go with the passive-aggressive advice "let pi build it for you". :-)
To be more constructive, what I did (as an experiencd SWE but a complete noob to agentic coding): went to pi.dev's extension marketplace and looked into all the new shiny stuff. Subagents, mcps, context and memory optimizers, skills. Using the most popular ones (not necessarily the best ones)
It was like 15years ago learning the new mindset of vim (and spending a ton of time to customize it to my workflow). My understanding is that Claude and opencode doesn't give you this flexibility.
Learning all these stuff drove me to also set up openwebui, and it was such a successful private project that I implemented it at work (with jira/confluence/bazel query access) and management said "we need this by tomorrow".
I believe the models matter not that much anymore. The "harness" does. (unless you just want to vibe code. Thebn, throw crap at fable and call it a day)
javier123454321 7 hours ago
peperunas 12 hours ago
rurban 12 hours ago
Happy with oh-my-pi (omp)
peperunas 12 hours ago
sergiotapia 5 hours ago
I've had a lot of fun with https://omp.sh/ - consider it like zsh -> ohmyzsh
Lots of sensible defaults and good tweaks/settings you just don't worry about.
try-working 12 hours ago
pi-role-model
kzrdude 11 hours ago
Do you notice any improvements with this update?
kmarc 10 hours ago
Not sure, haven't tried it today.
Last night it single handedly implemented a feature after a grilling session, and came back with the red-yellow-green risk assessment points that I mostly saw with anthropic models. I had to check if I'm using the right model, but it was DS4Flash.
So maybe I was using it already?
the_lucifer 9 hours ago
rdsubhas 11 hours ago
How can one set topp and temperature in pi?
embedding-shape 11 hours ago
You build it as an extension or create your own fork/copy of pi and add it. This basically goes for most things in pi, except the most basic stuff. It's basically meant for people who think "I'll just add that myself" rather than expecting it to be there out of the box or finding other's solution to it, for better or worse.
wkcheng 13 hours ago
If the benchmarks are real and reflect actual use, then this is an insane model. This 300B model outperforms the previous DS4 Pro preview model (1.8T params), and it looks like it outperforms GPT 5.6 Luna too. And it's still cheaper than Luna, even with the price decrease.
Crazy.
gr_norm 3 hours ago
OpenAI must've known this was coming, hence the Luna price drop. This competition is amazing!
cbg0 11 hours ago
On DeepSWE Deepseek is 54.4% and Luna is 67%
benjiro29 10 hours ago
But on Terminal bench, its
* DS4 Flash: 82.7
* GPT 5.6 Luna: 75.7
For reference, that puts it on the third spot behind GPT 5.5 and Fable 5. For some reason GPT 5.6 Sol is not showing in the leaderboard. If it did, then DS4 Flash was number four.
The thing is, even if Luna is better in DeepSWE and has the 80% discount. DeepSeek is still cheaper.
--------------
DeepSeek V4 | Flash GPT-5.6 Luna (New)
--------------
Input (Cache Hit) $0.0028 $0.02
Input (Cache Miss) $0.14 $0.20
Output $0.28 $1.20
--------------
Both Luna and Flash are heavy on the reasoning > output. And the cache hitrate + prices also matter.
Reality is, you can not go wrong with Luna or Flash at those prices. And remember, DeepSeek V4 Pro is still in the rafters, what is ironically closer to Luna's new price.
flashblaze 8 hours ago
dannyw 11 hours ago
Deepseek and moonshot are the only two providers I consent to training for.
arizen 11 hours ago
Why not also Qwen?
dannyw 10 hours ago
Qwen/Alibaba have stopped doing open weights releases for a while. No grudge or anything, I'm certainly not going to look at a gift horse in the mouth, but both DeepSeek and Moonshot have been very consistent with open weights as well as sharing actually detailed research.
In terms of open research, China has absolutely overtaken the US.
anon373839 10 hours ago
siva7 6 hours ago
CCP will be happy! Go on and share all your data with them..
neya 4 hours ago
Yeah, because, your data is best collected by the US labs, right?
US or American, don't trust anyone, these are open weight models. Host them yourself if you feel strongly about privacy (as you should). I honestly don't know of any OSS open weight models from the US labs as good as Kimi K3 or Deepseek V4 though.
fdsjgfklsfd 5 hours ago
Yes, they trained on my open source code and Wikipedia edits and Stack Overflow answers that I shared freely, so it's only fair that they release their models as open source and share back to the community that created them. I only allow my training data go to open source models.
gr_norm 3 hours ago
chorizo 5 hours ago
I’m developing open source tools, so happy to have all my context traces be public if it helps improve future models.
anigbrowl 2 hours ago
I will! I pay their sales too and consider it quite reasonable.
genxy 3 hours ago
At least you get based models back at some point. You supply data, they supply compute and open weights.
bel8 2 hours ago
Yes I will tyvm.
They release the models back for free.
applicative 8 hours ago
Xi going to shut down open-weighting of them in a matter of months. No one seriously doubts this. They will be too powerful and they will be gone.
satvikpendem 6 hours ago
He explicitly said he wants open weight models at his recent speech at an AI conference in Shanghai: https://news.ycombinator.com/item?id=48970449#48970784
applicative 2 hours ago
gravypod 8 hours ago
What would the benefit of this be? If China stops open weights, US labs still have the intelligence frontier. Maybe once Chinese models have speed, cost, and intelligence beat but right now they don't.
Undermining out entire economy by subsidizing the release of DIY versions of our main economic drive sounds like a huge win for China.
applicative 7 hours ago
anon373839 8 hours ago
Reuters was reporting that rumor. And then Xi made a public appearance at a conference in Shanghai where he said the opposite of that rumor.
applicative 7 hours ago
GTP 7 hours ago
We can only wait to see if this is true. Another possibility could be that it was an answer to USA's government considering a ban on chinese models. In this way, he fueled the discussion around the importance of open weight models.
applicative 7 hours ago
darkwater 8 hours ago
What makes you state this?
riskd 8 hours ago
baublet 8 hours ago
codemk8 2 hours ago
Me.
1234letshaveatw 7 hours ago
That isn't the Chinese way. They are much more focused on undermine and extinguish. Just look at the European car industry- on its way to being non-existent after the market was flooded, bye bye manufacturing base. Undercut the US AI providers and wait them out, they will go private after they have a stranglehold
bwfan123 5 hours ago
applicative 7 hours ago
ReptileMan 8 hours ago
It is true. And only Chinese Han people can continue work on what is already released. But they will be forbidden. /s
Sarcasm aside - if the community can't get their shit together to continue improving what is currently public, well - we don't deserve free stuff and open weights.
heyalexej an hour ago
Long story, I have humongous zai GLM 5.2 token budget that I'm using in a similar fashion as many comments explain here. GPT 5.6 or Fable 5 for planning, GLM for implementing, researching, extracting and many other tasks I consider grunt work. Very happy with the performance, speed isn't all that good though. I'd be curious to hear from someone who works with both, DeepSeek and GLM side by side.
ggcr 13 hours ago
Woah, a 200B model competing with GLM-5.2 and getting close to Opus 4.8. Quite impressive.
If those numbers translate well to its general capabilities, with the great caching DeepSeek has, I feel like this model will get tons of usage.
lostmsu 9 hours ago
Not just 200B model, it is only 160GiB.
ignoramous 8 hours ago
MiniMax's another lab that's known for relatively smaller models (their latest, M3 is 295b) that punch way above its weight.
Goranek 13 hours ago
Kimi K3 (instead of Opus) for expensive stuff, DSV4 Flash for tasks (instead of Sonnet)?
Does this make sense?
baalimago 12 hours ago
Looks like they will release an updated version of deepseek-v4-pro soon, which most likely will beat kimi k3 at both intelligence and cost (judging by the vast improvements to dsv4 flash)
geek_at 13 hours ago
it does! I'm always amazed when I use DSV4 flash for coding or server checks and after an hour of working with it my (pure api call) balance is about 30 cents
wg0 2 hours ago
Don't know about the bench marks but I am getting Opus 4.7 level performance at fraction of cost with DeepSeek V4 Flash set to high. It is a reliable workhorse.
thirtygeo 13 hours ago
For both US and China models - what standard security checks and QaQc are you all doing? We're running small gamuts to test for unsolicited jailbreaks (model jailbreaks you) and incorrect records (Fake Accuracy - as Easter Egg or common thread) meaning falsified logic or information cooked in by the developers, rather than the training data speaking for itself
baalimago 13 hours ago
Very promising. So it will both keep the speed and reduced price, yet exceed performance of the quite sufficient deepseek-v4-pro?
Should be extending the lead in intelligence/cost index, as deepseek-v4-flash already were the most price efficient model, which now becomes even better. Although, in the deepseek APIs, the cost is leaking all information about codebases to China.
ilaksh 9 hours ago
Supposedly better than GLM 5.2 according to at least one benchmark.
namuol 3 hours ago
Can someone please explain how these models aren’t just fine tuned for benchmarks? I’m not plugged in to this space much but it seems like such an obvious problem…
sroerick 3 hours ago
They definitely are - but also people are using them pretty extensively for work. So ultimately you can't really fake "is it good". But there's no real measurements of that when a model is released, so we are stuck with benchmarks.
namuol 8 minutes ago
I will continue to ignore the benchmarks.
sim04ful 12 hours ago
Why didn't they increment the version as atleast a patch update
bel8 2 hours ago
So whoever is using DeepSeek V4 flash gets a massive upgrade without having to change any config.
crvdgc 7 hours ago
My guess would be the version number tracks the architecture change, not the weight change.
rubslopes 9 hours ago
I was also so annoyed by this. It makes communication all around difficult. Why don't they simply call it 4.1 flash?
kzrdude 11 hours ago
Yes, and even if they seem to have updated endpoints in place they should assign the service a version number.
markasoftware 12 hours ago
They want to be the next Google it seems
yewenjie 10 hours ago
What was preventing them from calling it v4.1-Flash to distinguish it better?
lionkor 8 hours ago
Sounds like it'll replace v4-flash, v4.1 would be nice to keep both available. On the other hand, it's nice to just get an improvement on anything that asks for "deepseek-v4-flash" without having to change the model string.
krapht 8 hours ago
I think that's backwards. Anything that changes the performance of a model deserves a minor version bump. A new model has to be qualified before being pushed to production; but we don't get the choice here, just cross your fingers there are no regressions at all on all possible tasks the model might be asked to do.
xbmcuser 5 minutes ago
halJordan 4 hours ago
lionkor 8 hours ago
halJordan 4 hours ago
It went from "ds v4 preview" to "ds v4". That's enough of a distinction.
troglodytetrain 2 hours ago
This is very exciting, my own niche micro-saas has already been able to make heavy use of DeepSeek-V4-Flash for my use case, I am looking forward to seeing how performance improves.
throwa356262 7 hours ago
Gentlemen, start your DGX Sparks
alecsm 8 hours ago
I've been using DeepSeek Pro for a while and Flash only for certain dumb tasks where I only need the speed of a LLM and not big brains.
I find the newest OpenAI and Anthropic models to be way better for big tasks that require many decisions but I don't like that anyway because I lose track of what's being done.
Knowing what I want for every prompt makes DeepSeek Pro the best LLM for me. It allows me to work relatively fast at a very low price.
sqemo 11 hours ago
DeepSeek is great for tasks and software I already know well. Even if it gets something wrong, I can usually verify it myself. But when I'm working with a programming language I'm not familiar with, I prefer using Codex or Claude.
k__ 10 hours ago
Yeah, it needs quite some hand holding.
I didn't do much agent coding and had a mix experience.
1. It would build something that was in the spirit of what I wanted, but unusable in practice.
2. It would build something quite useful, but only the public APIs were nice, the deeper code layers would get more and more convoluted.
3. It would built what I wanted and it would have okay-ish code.
However, for 3. I also had to add a custom AGENTS.md, many more code example, extra repos as subtrees, and review any code that had new concepts.
Much more work, but still much less than typing it all by hand.
arjie 13 hours ago
Oh my goodness what an update. I need these weights. It's an incredible model for the size. The improved tool calling etc. should be able to make my harness way simpler. This runs at mega-speed on prosumer hardware (2x RTX Pro 6000).
amunozo 13 minutes ago
How many tok/s are we talking about?
mordae 12 hours ago
I was just using it when it landed. It started reasoning more extensively from nowhere and precision went up a lot. It also changed its prose style for the better. Looking forward to weights.
nickandbro 13 hours ago
I wouldn't doubt GPT 4.6 Luna being in the top left quadrant's center on the Cost per Intelligence Index is not concerning for Liang Wenfeng. You have to remember DeepSeek v4 flash even though a bit cheaper, does not have vision abilities, which is a big draw for agentic tasks.
I admire DeepSeek's openness, but even they have been raising prices after their discounts.
minraws 13 hours ago
They haven't raised prices though the plan was to increase it with peak hour usage for V4 Pro GA release, they didn't do that, so no price increases there I believe.
As for vision yeah it sucks but Luna is also 2x input and 1.5x output for 1M context...
That's around 0.4 in/1.8 out
DSv4 is wayyy cheaper.
And it's open now you have Luna at home if you have a decent set of GPUs you can run this on 2Sparks or one very expensive Mac or just like 6-8 5090s..
minraws 13 hours ago
I know most folks can't afford it I am working on making it viable to rent shared hosting the biggest issue is data leak and prompt injection attacks with shared hosting. (Since the server owner connects to your main system via the coding agent)
I guess using a ZDR provider is good enough for now.
dudisubekti 13 hours ago
Gpt 5.6 Luna cache read is $0.02 per mtok
V4 flash cache read is $0.0028 per mtok
That's not "a bit cheaper", just saying
nickandbro 13 hours ago
That's a good point. Yeah their caching input is insane.
ignoramous 8 hours ago
gpugreg 10 hours ago
According to the leaked call transcript, DeepSeek is working on vision for V4. Not sure when it will land though.
re-thc 13 hours ago
> I wouldn't doubt GPT 4.6 Luna being in the top left quadrant's center on the Cost per Intelligence Index is not concerning for Liang Wenfeng.
The leaked interview has him saying it doesn't matter... as much as open source doesn't matter. There's enough in it for everyone right now and they aren't after everything.
Perspective: DeepSeek doesn't have enough infrastructure to serve their target customers already.
nickandbro 13 hours ago
Can you post the link to the leaked interview? From what I understand he has been pretty tight lipped for a guy who has a larger stake worth more in his company than Dario does in Anthropic.
ifwinterco 8 hours ago
Rzor 12 hours ago
amelius 10 hours ago
Note: if you are having success with a model, then please post what you are using it for. Writing HTML/CSS is very different from writing Rust/C++ or doing maths.
fdsjgfklsfd 5 hours ago
I use DeepSeek V4 Flash (before this update) for most things:
- OpenCode for codebase editing: python scientific computing and LLM projects
- Open Interpreter Classic (python version) for Swiss army knife terminal replacement one-off task type stuff.
f311a 10 hours ago
I do Python, Go and Rust. Go works the best, I would say Python is worse than Go.
Rust works perfectly fine, but when I use Rust, I usually pay closer attention to performance, so I have to guide it a bit, to improve cache locality, use simd, avoid unnecessary allocations and so on. Terra has the same issues with Rust. If you always prompt models to achieve the best performance, the code is usually no the one that I want, they optimize unnecessary/cold parts or blindly optimize stuff where compiler takes care of the optimizations already.
See my other comment for more information on how I work with it. In short, keep the changes under 1k lines, context under 120k (ask it to use subagents), drive the architecture yourself.
sparse-Matrix 10 hours ago
I've been having better-than-most performance using qwen3.6 to write python/flask.
I tried using it to generate some rust code yesterday, and it generated much code but only ever came within 1 error of a testable build. The 4th or 5th full rewrite is sitting in the buffer right now.
I'm currently looking to up my game with Bottlecap AI's return of qwen3.6, 'thinking cap'.
Alleged to be twice as fast and superior at coding over extended sessions (vs. 3.6).
We'll soon see.
ilaksh 9 hours ago
You should specify which model size and quant because there are many of both.
bel8 3 hours ago
flash for code:
c#, TypeScript, PHP, SQL, CSS, HTML.
also features, tests, fixes, refactoring and planning.
it's super fast, smart and dirt cheap.
dandaka 8 hours ago
Text classification, structured data extraction, rewrite
Reubend 13 hours ago
They're always very understated in their update descriptions. This is actually a HUGE improvement in the model's capabilities rather than just a small tweak.
f6v 12 hours ago
I'm thinking of using ChatGPT for making plans and V4-Flash for execution. Does anyone have good advice on pairing different models?
davidjade 5 hours ago
I stumbled into this same workflow idea. I hadn't really used anything other that Chat GPT Sol (medium) and when I wanted to start using code agents, I just asked GPT how to start. This evolved into having GPT do the brainstorming and writing the initial prompt for a project feature/change to hand over to Codex working in Sol light mode. These prompts where often a couple pages long as they had a lot of the architecture details worked out already. Codex would then come up with a plan and I'd take that back to GPT to review. GPT would make some suggestions, which I'd throw back to the agent and off it went.
I got super great results working this way. Maybe there is a better way to integrate the two modes though. But my first real project was end-to-end 100% working correctly from the first run - about 10,000 lines (including tests) of greenfield code. I did have it work in manageable chunks that I could easily review - not one-shotting the whole thing.
Now I'm thinking about plugging Deepseek into Codex to be the coding model and see how it goes.
embedding-shape 12 hours ago
Pairing non-APIs with APIs tend to be a hassle, and risky, as usually that breaks the ToC. I guess easiest for you to test if it's worth using the OpenAI API, is to manually copy-paste responses between wherever you run V4-Flash and the ChatGPT UI. What I've done in the past is basically .zip up the entire project directory, ignoring files from .gitignore, then asking ChatGPT Pro to inspect that and come up with a plan, then you paste that to where you have V4-Flash. Basically how we did "vibe pair programming" before the TUI agent harnesses appeared in the ecosystem :)
yuzuquat 11 hours ago
reminds me back in the day when we used to do version control with google drive. you generally don't need to do this anymore i don't think. it should be fairly trivial to open the codex cli in the project directory, ask it to do some analysis and planning -> save to md. then opencode with deepseek to read that md and start implementing. consider that some cross-model subagent workflows already exist: i use claude/fable to plan and there are openai plugins to directly spin up codex subagents. this latter approach gives claude/fable more directly control to babysit the codex agents whereas the former is more of a clearcut handoff
embedding-shape 11 hours ago
cthulberg 11 hours ago
https://github.com/obra/superpowers
I use Opus/Sol with for /brainstorming, deepseek (on Pi) for /subagent-driven-development
I love it, the docs are easily editable and when I'm ready Deepseek is faster/cheaper then everything on my coding plans.
leobg 10 hours ago
Source reads like Tony Robbins for LLMs:
Excuse
"Too simple to test"
Reality
Simple code breaks. Test takes 30 seconds.ilaksh 9 hours ago
I wonder when the antirez/ds4 group will have an update to their high accuracy 2 bit quant.
Although it's funny that I am thinking about that at all because I have a 2060 :P . My local inference is playing with Gemma 4 E2B and MiniCPM 5 1B.
jtbaker an hour ago
Looks like a couple of people are already on it: https://github.com/antirez/ds4/issues/635
throwdbaaway 9 hours ago
Objectively speaking, the 2 bit quant from antirez has very low accuracy. Meanwhile, his 4 bit quant does have decent accuracy, but is a bit pointless by being bigger than the full precision MXFP4 quant. Anyway, they all work fine in practice.
Tepix 10 hours ago
Sounds like a big improvement.
No mention of weights, just API. When will the updated weights be released?
fdsjgfklsfd 5 hours ago
Was already released before your comment: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
miyuru 13 hours ago
Judging by the openrouter leaderboard ranking for today, it looks like Dv4F us more popular than mimov2.5.
https://openrouter.ai/rankings?view=day#leaderboard-table
These days cost per task is more important, and SOTA models have become expensive.
benjiro29 10 hours ago
MiMo, the overlooked sidekick to the hero. Will be interesting to see what Xiaomi bring to the table.
These massive jumps in cheap models, is really great times!
KronisLV 13 hours ago
Wonder how good the proper version of V4 Pro will be.
I'm still considering pulling the trigger on the annual subscription of Kimi for K3 but it's sometimes slower than I'd like (at least when compared to Anthropic) even on their Vivace plan, and the token limits on the GLM Coding subscription for GLM 5.2 were too easy to hit.
HyperL0gi 9 hours ago
Is anyone using DSv4 for their agents that is not related to writing code? Curious about use cases specially for someone using gpt-5.4 mini for classification, categorization, etc
freakynit 8 hours ago
I use it to conduct thorough online researches. Plug-in some online search MCP (like the one I use: https://jerrysniffs.online ), and the flash models dig through the internet for dirt cheap..
Most of the times, the total cost, including search API's, is less than $0.05 for full deeply researched output, and the research is actually good.
Lalabadie 7 hours ago
It's been good for one-off cases in my limited experience. I would describe its behaviour as Sonnet-shaped, if that makes sense to you. Good answers but it often decides to reason a lot about simple things before getting to an output.
At the speed Flash has on most providers, it doesn't really turn into a latency concern.
markab21 7 hours ago
We use it at a moderate scale, self-hosted on B300 hardware. It's great :D
QA analysis of voice transcriptions. Napkin math: we operate at 2-5% of the cost of running on Equiv Frontier, though this changes near-weekly because pricing is so volatile.
It took us about a month to get the inference configured to achieve these numbers. But if you can get your hands on a pair of B300 GPUs and the context works, it's untouchable for price/performance.
(B200 would work, but you don't have the B300's memory, which lets you run it on 2xGPU instead of 4xGPU... with Dspark, it's like magic)
On a side note, for tasks that don't require the intelligence of DS v4 flash, we're using Nemotron-3-super with incredible success. I'm shocked we're not seeing more adoption of this model, given how easy it is to fine-tune and how blisteringly fast the nvfp4 version is. (A single B200 GPU can produce an insane amount of throughput with Nemotron 3 Super.)
w2seraph 2 hours ago
This made my day !
vladukha 12 hours ago
Where do you guys get deepseek? I'm hearing a lot of good reviews and want to try it with my pi config. from the deeepseek themselves, openrouter, or anywhere else? does it make a difference? [edit]: whoa it is really fast. will take some time to evaluate quality thou
u8080 10 hours ago
Directly here: https://platform.deepseek.com/ Easy top-up and pay as you go. Availability and speed are very good and it is the cheaper option.
psibi 12 hours ago
I've been using DeepSeek directly. I've heard from colleagues that using it via OpenRouter is slower, but I'm not so sure about that.
ticoombs 12 hours ago
> does it make a difference
Probably not.
But Opencode-Go is a great solution for those who don't want to pay DeepSeek directly (or can't due to reasons)
Selfish referral code: https://opencode.ai/go?ref=R1AJZT4VBX
embedding-shape 12 hours ago
For hosted APIs, it's a lot cheaper to use their own infra, caching seems a hell of a lot better there compared to OpenRouter, and indeed the tok/s seems higher. They also have peak/off-peak pricing, so if you can hold off with your request, you get a pretty big discount.
Otherwise, if you're trying to run it locally, even really low quantizations like DeepSeek-V4-Flash-IQ2XXS-w2Q2K-AProjQ8-SExpQ8-OutQ8-chat-v2-imatrix seem to actually not be so dumb compared to smaller models with same quantization, might be worth a try if you're sitting on a lot of RAM/VRAM yet not industry-scale amount :)
chronogram 10 hours ago
I use it directly: https://platform.deepseek.com/usage
3rd party providers on OpenRouter can be cheaper but it's already so cheap.
kzrdude 11 hours ago
Opencode-go gives you $60 worth of DS V4 api usage for $10 per month. Right now I think it's hard to exhaust that when using flash exclusively, and plain API use might even be cheaper! Anyway, for DS usage it's a good deal.
gpugreg 10 hours ago
To add to this, the $60 only applies to DeepSeek-V4-Flash and a few other models. For DeepSeek-V4-Pro, the amount is $15.
https://opencode.ai/docs/go/#usage-limits
Previously, OpenCode Go had higher API prices for some models, but now they lowered the API price and simultaneously reduced the allowance.
kzrdude 10 hours ago
Lalabadie 10 hours ago
Opencode also have a ZDR (zero data retention) deal with them – if I recall correctly, that's not something you can enable as an individual DeepSeek subscriber.
anon373839 10 hours ago
gpugreg 10 hours ago
lucianmarin 8 hours ago
OpenCode harness gets the most out of DS V4 Flash model. You can implement any coding task, fast and cheap.
Gigachad 12 hours ago
I used it through openrouter. Plugged in to the vs code copilot bring your own key thing.
Played around for a few hours and used up 80 cents of tokens.
Lalabadie 10 hours ago
Anecdotal data from my own tests: allow only one provider if you want good cache usage on OR. 80 cents is probably 4x the price you should have paid.
Providers' cache hit stats are available to consult, and only 1-2 of them behave properly if I remember correctly, zero if you request providers that don't store and train on sessions.
darkest_ruby 11 hours ago
Openrouter
wolttam 13 hours ago
Hooray! This model makes me very optimistic about the future of local inference. The CyberGym score stands out to me.
egeozcan 13 hours ago
Every time I want to have fun coding something with natural language processing, I use deepseek flash. It's just incredible for the price. I have a fairly popular app with 400 users that uses DeepSeek in the background and it still didn't hit even 50 bucks of usage in a month.
indigodaddy 9 hours ago
Mind sharing the name? Sounds interesting.
PhilippGille 13 hours ago
The previous V4 version wasn't called “Preview” by most inference providers. For example, the OpenRouter model slug was `deepseek/deepseek-v4-flash`. So now there will be confusion when someone talks about V4 Flash or when someone offers V4 Flash inference.
Why not call it V4.1?
petu 12 hours ago
It probably would be called 'deepseek-v4-flash-0731' in API
edit: nope, at least deepseek kept "deepseek-v4-flash" and just updated model underneath. I guess preview is no longer worth serving with that release and you'd have to look through inference provider docs to see if they've updated, yeah..
PhilippGille 10 hours ago
That's what I mean. On DeepSeek it's now just `deepseek-v4-flash`, while OpenRouter calls it `deepseek/deepseek-v4-flash-0731`, so now when someone talks about DeepSeek V4 Flash, like in benchmarks, or other inference providers, which version do they actually mean?
The `-0731` style suffix is worse compared to a proper version bump like V4.1.
Macuyiko 10 hours ago
kzrdude 11 hours ago
Does it still say that it's an anthropic model, when asked? I would guess new post-training has fixed that.
petu 11 hours ago
bermudi 11 hours ago
DeepSeek being DeepSeek. v3 and R1 went over the same and had multiple versions
try-working 13 hours ago
DeepSeek themselves called it `deepseek/deepseek-v4-flash`. Pro is still like that.
PhilippGille 10 hours ago
Yes that's my point. The old and the new version are different in capabilities, but now when someone talks about DeepSeek V4 Flash (in benchmarks, on inference providers), you don't know which exact version it's about.
Some providers like OpenRouter now call it `deepseek-v4-flash-0731`, but even in places like here on HackerNews people say things like "Sonnet is better than DeepSeek" without specifying a version or a reasoning effort, certainly no one will mention that `-0731` suffix when talking about DeepSeek V4 Flash.
benjiro29 10 hours ago
flysoft 13 hours ago
Finally have a model with usable intelligence, at a reasonable price. Can't imagine what Pro GA would look like, considering pro preview has only 1.6t parameters.
throwaw12 11 hours ago
how different is their harness from Pi coding agent harness, is it possible to make an extension for Pi which can implement deepseek harness?
storywatch 13 hours ago
How's their performance in English prose? We are currently searching for cost effective ways to keep story wikis up to date.
amunozo 12 hours ago
I am interested in this too, as I think it could be these models are overly optimized for coding. Let me know if you figure it out!
nathaah3 12 hours ago
DS v4 flash has been my goto model for tasks in work. its been unsurprisingly fast and cheap.
k__ 10 hours ago
I'd take more throughput while everything else stays the same.
kamikazechaser 13 hours ago
The flash variant is on par with Sonnet 5 on DeepSWE (54%). Big, if true.
znnajdla 10 hours ago
The conspiracy theorist in me wants to think that the 80% drop in GPT 5.6 Luna prices today is correlated with this update from DeepSeek. Perhaps OpenAI has already hacked its competitors with it's Mythos-like models and is aware of what competitors are doing and is able to react in advance.
nchmy 9 hours ago
Seems to me that it's the reverse. Open ai dropped prices to compete with deepseek etc and then deepseek released this update to negate Luna.
spwa4 14 hours ago
In case people want to run it, it's DeepSeek-V4-Flash-284B-A13B. So it should just barely run on a single B300, and it's small enough that it'll barely run on an M5 Max too.
wolttam 14 hours ago
It runs really well on 2 DGX Sparks - 60t/s
Tepix 7 hours ago
Yes, the Dual DGX Spark looks like the sweet spot for this model for now. Good preprocessing speed. Lots of context. Fast enough for 1-5 devs perhaps. Around 8200€ as of today (used to be 6000€).
Dual Strix Halo is much slower and current Macs with 256GB are both slower and more expensive (Mac Studio M3 Ultra 256GB around 12000€).
To get something faster than the two Sparks you'd need to spend more than $22000 for a server with 2x RTX Pro 6000 at $10000 each.
Beyond that you could get 2x AMD MI350P.
arjie 13 hours ago
Not yet, right? That's the old DS V4 preview release. We're still waiting for the weights to come out.
benjiro29 10 hours ago
Probably the same. When the same base model is trained, the weight do not tend to change a lot. GLM 5.0 > 5.1 > 5.2 are the same base model, that just kept being trained. Weights hardly change as a result. Think in the like few percentage points size difference.
lukan 13 hours ago
"it'll barely run on an M5 Max "
The max version I could order now with 128 GB?
If so, the price for local inference would be 12 000 € vs 500 000 € for a B300.
NitpickLawyer 13 hours ago
There's also the 2x spark way, which should be ~8k eur? Someone down the thread reported ~60tps for 2x sparks. That's totally usable for local inference.
You can also do 2x 6kPRO in a workstation, for ~20k.
matrik 12 hours ago
spwa4 12 hours ago
reverius42 12 hours ago
I'm running a useful quantization of the previous version of Deepseek-V4-Flash -- quite well but with so much fan noise -- on a MacBook Pro M5 Max with 128 GB.
spwa4 12 hours ago
500k is for the 8x B300 version. Which is the only one you can buy atm. But technically a B300 card is more like 60k, just impossible to get.
XCSme 9 hours ago
Can't really use it now, without giving away your data:
> Trains: this provider may use prompts for training and may retain prompt data.
Philpax 8 hours ago
That will cease to be a problem in the next 24 hours, now that the weights are out: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
ra 12 hours ago
What's the best way to run this on a 64GB M2 Pro?
petu 11 hours ago
Weights are yet to be released (maybe in 24H, Deepseek has track record of releasing same day).
https://github.com/antirez/ds4 is often mentioned for DS4F on Mac, but 64GB is likely not enough to achieve reasonable speeds (official weights should be ~160GB).
Tepix 7 hours ago
DS4Flash has 284B weights. 64GB? No go.
zozbot234 6 hours ago
The native weights are 4-bit for the sparse experts, and they quantize to ~80GB with limited degradation in real-world performance. That's a viable target for 64GB with SSD streaming, though it will be slower than keeping the whole thing in RAM (especially on a M2-class machine with its slower storage).
sparse-Matrix 10 hours ago
This may come as a surprise to a lot of AI concerns, but I have -zero- interest in paying for a model.
kzrdude 9 hours ago
Then DS V4 flash is pretty relevant, because it's pushing prices down.. as well as being self hostable with a large enough rig.
Tepix 9 hours ago
Why mention it? Just download the weights when they become available.
sreekanth850 8 hours ago
how this compare to luna high with reduced pricing.
mekky16 10 hours ago
if they were anthropic they wouldve just released it as a new model
truth_seeker 11 hours ago
The magic of post training with valuable dataset
sourcecodeplz 12 hours ago
i've made a comparison between this and GPT Luna (recent %80 price drop)
https://x.com/SourceCodeplz/status/2083099712760987746
i prefer GPT-5.6 Luna honestly
dakolli 10 hours ago
Chinese labs rushing to release models this week, because it's inevitable that Washington regulates Chinese models in the next 4 weeks. All the US AI leaders have been taking trips to Washington this week, what do you think they're there for..
dnhkng 14 hours ago
DeepSeek V4 Flash (Preview → 2026-07-31)
• Terminal Bench: 56.9 → 82.7 (+25.8)
• Toolathlon: 51.8 → 70.3 (+18.5)
Compared to GPT-5.6 Terra:
• Terminal Bench: Flash 82.7 vs Terra 78.4
• Toolathlon: Flash 70.3 vs Terra 53.1
• DeepSWE: Flash 54.4 vs Terra 69.6
• Agents' Last Exam: Flash 25.2 vs Terra 50.4
Trading blows with Terra, which is pretty interesting. No clear winner on these benchmarks, and wildy differeing scores. Very interesting!
villish 13 hours ago
benjiro29 9 hours ago
https://www.tbench.ai/leaderboard/terminal-bench/2.1
> 78.4
The real score is always the official benchmark.
We need to see later if DS4 flash 0731 is going to maintain the score but we need to look at the official benchmarks.
Already seen a PR for DeepSWE to update the benchmark with 0731, so we can verify claimed vs official.
villish 5 hours ago
throwaw12 13 hours ago
Open flash model is competing against OpenAI's 'Sonnet' model at the price of GPT 3, I am really excited about this release, hopefully it holds up in real work as well
bayesianbot 12 hours ago
IIRC GPT 3 was priced at per 1k tokens, had to check, the biggest GPT 3 model from OpenAI was $0.06/1k, so $60 / 1M. gpt-3.5-turbo was the first model after ChatGPT and that was $2 / 1M. And no caching. So not really in the same ballpark
Iolaum 14 hours ago
Since they did this with their own harness I m not sure it's apples to apples comparison.
NitpickLawyer 14 hours ago
> not sure it's apples to apples comparison.
They're literally comparing the previous version of the same model with the new one. It's based on the same architecture, same pre-trained model, just different post-training. It doesn't get more apples to apples than this.
dnhkng 14 hours ago
dnhkng 14 hours ago
It will be fair if they release the harness though. I think now the future will be paired model-harness releases, not just weight dumps.
The performance changes are so big with the right harness that is makes sense to engineer the harness and fine-tune the model to one another from the start.
yms_hi 14 hours ago
I think it's better than GPT Luna.
try-working 13 hours ago
Let's see how the market reacts.
Havoc 12 hours ago
>benchmark results far exceeding V4-Pro-Preview:
Wow that's crazy
Good times for those that don't need strict data protection
blackoil 12 hours ago
These being open provide much better data protection.
kzrdude 11 hours ago
Open weights means that there is a free market on providing inference using this model. Some of them will offer good data protection. (Well, hopefully)