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.
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.
Qwen has a pinned tweet stating that 3.8 will be released as open weights soon. I guess it remains to be seen, though, if they’ll do the smaller model sizes or only the big 2.4T one.
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.
> Maybe once Chinese models have speed, cost, and intelligence beat but right now they don't.
Yes this is why I referred to 'months' and was downvoted by people who can't distinguish their politics from reality. You are restating exactly the text you are criticizing.
This is the nature of mechanical parrotlike repetition of propaganda:
a) You can have 'frontier' models with closed weights. Your sentence is basically a contraction within itself, again, the weight of ideology.
b) No one actually knows whether or not the closed weights of Bytedance are beyond all existing frontiers. Again the weight of ideology blinds you to the fact that the real 800 lb gorilla of Chinese Ai is more closed than Anthropic.
The latest ByteDance Seed model is behind Opus 4.6 in performance. It’s not even in the conversation atm.
The parent made a totally coherent commment that China closing off their open weight models would cede the frontier (and majority of AI users worldwide) to the USA. Nothing you said counters that point.
Do you have a source for that? The NYT article does not quote Xi making that statement, and I can't find a source for that. In fact, Xi delivered veiled criticism of the security posturing the US has done:
> We should put in place laws and regulations, technological monitoring, early warning and emergency response systems in order to strengthen the line of security, prevent abuses and malicious use, and ensure that AI is always under human control. In the meantime, we should jointly oppose overstretching the national security concept in the field of AI and placing one country’s security over that of others.
There was a blog post by a state-linked broadcaster cited in the NYT piece:
> “China supports openness, but this does not mean it advocates for the unconditional proliferation of all capabilities,” the blog said.
The article also quoted an American journalist:
> “If these models do reach those dangerous capabilities, they are not going to let it be a free-for-all in terms of releasing them," Mr. Sheehan said.
But a distinction has to be drawn between real dangers (which many people believe LLMs have not actually shown, to date) versus "dangers" hyped up marketing purposes or domestic regulatory-capture motives. Presumably the Chinese government is less interested in the latter.
But the speech doesn't say that? I'm looking at the full text. For example:
> We should take seriously the various types of inherent and secondary risks that AI may trigger.
The risk portion of the speech is focused more on application-level risks than model capabilities. Which is a much more sensible regulatory framing than what Silicon Valley has proposed.
replying the request for sources with "as xi stated" without bringing the sources again is quite misleading. It seems you have nothing to back what you are saying.
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.
Xi actually said that models of the strength to pose security issues will be permanent state secrets -- in the same speech that the first western takes translated as actually using the words 'open weights' which of course nowhere appeared. Now: when will "models of the strength to pose security issues" appear? Maybe my inference 'months' is wrong, but they seem to be advancing rapidly.
The administration blather about banning open weights is characteristically confused. Xi has already stated (what is obvious) that he will ban security-endangering weights and keep them a state secret.
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
Even Xi's speech was as paranoid as Dario Amodei about the possibility of Chinese AI achieving something at the level of say Mythos. If you seriously believe such a thing will be on Hugging Face I really don't know what to say.
There is already something on HuggingFace at the level of Mythos. It's called Kimi K3 and it's running laps around Opus5, Fable5, and Sol56 at cybersecurity.
It's so good that the US government is rushing to ban all Chinese models as fast as they can.
Fine, you believe Anthropic was lying about Mythos. In fact no view of the matter is needed to formulate the proposition above. It is clear China will soon have a model with the powers imputed to Mythos but which you deny of actual Mythos. It is a trial to deal with this level of ideological blindness; 'ideology has no outside'. If someone says 'when they get Mythos...', declare the possibility of Mythos to be the real lie. But it is quite possible and will be coming in months.
Reading between the lines, it's fairly clear that Mythos Preview only got its reported results in offensive cyber thanks to a highly specialized harness and humongous amounts of test-time compute. We also don't know what kind of post-training/fine-tuning it got on cyber, Anthropic only stated that it wasn't specifically trained for cyber-offense but that leaves a lot of scope for training on other related things. Overall, the Mythos story is a lot less interesting than most people might like to claim - but note that this doesn't require any actual lying.
The interest around K3 is a lot more defensible because we actually know what the architecture looks like, how much effort it takes to get it to run, and what kinds of results it gets on cyber evaluations. And no, it's nowhere near "dangerous" enough to where people might honestly want to ban it for real safety reasons. It does a good enough job at fixing cyber issues, but that's hardly a safety concern.
No, reading Xi's speech. They aren't going to give their Mythos - presumably a few months away - to the Sinaloa cartel or Uighur hackers or the US military or etc etc
They have already released their models (GLM 5.2 and K3). This is why Amodei and the rest of the effective altruist lobbies are trying to pressure the US government to both ban open-source models and pressure China to do the same.
I’m curious to see why you’re so eager to ban open-source LLMs while being okay with Anthropic’s control. Do you think they’ll rebel against you?
Wow! Amazing observation! Wait, do you think there could have been any pro Chinese propaganda? Perhaps even some taking place at the present time? I think it could be possible?
I'm sure there's plenty, but if you put on your rational hat and think about the motivations of the Chinese government you'll realize this claim makes no sense, and there's plenty of evidence and good reason saying the opposite
I'm a) quoting Xi's speech and b) projecting rapid China development to Mythos level. Either you think China will never get to 'Mythos level' because you think China bad; or you think they will release the weights, because you think China bad. Its incredible the weight of ideology over facts in this discourse
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.
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.
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.
> 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?
How so? The code and reasoning patterns certainly match, so does the intelligence level. I probably wouldn't know if they're secretly serving Luna instead of Sol, but it's definitely an OpenAI model.
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.
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.
Openrouter has cheaper inference than deepseek for the prior version of v4 flash, which I suspect will happen within a day or two with this version, and they don’t offer subscriptions. Are you confusing it with opencode?
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.
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).
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.
Aside from what squidbeak already said, I have to say the quality varies a lot, and goes above the "good" threshold enough times that using these models is worth the time and effort. Compared to something like GPT-5.4 to GPT-5.6 Codex models, they cost 50-100x less (fifty to one hundred TIMES less), and can do most annoying or well-specified tasks just as well.
The main difference is that deepseek is bad at prose, and Codex models are much more eager to use tools provided to them (which is usually fine, they often make like 3 todos via tool calls for a simple one step task which is silly though), and deepseek in general benefits from good instructions more than GPT models maybe do.
Deepseek becomes much better if you give it tools for asking clarifying questions, doing self-review with subagents, and so on. The more tools it uses, the better the signal-to-noise ratio in its context, and the more consistent the output.
You must still review 100% of the code as if it's trying to sell you insurance.
An experienced software engineer (read his profile) praises the value he's found in Deepseek, and gives some real data showing how affordable that value is.
Then you, dakolli - out of generosity and minute-to-minute devotion to enlightenment - sacrifice time from your busy day to sit down (though perhaps that's been painful lately?) or stand up with your phone - and offer a profound, deeply thought-out counterpoint in the following form (and I'll paraphrase):
How one is affected psychologically by LLMs 'hallucinating' presumably closely tracks how one is affected by pathological liars. Most of the things pathological liars say is true and if you can put up with it, one can work with them and learn from them. But some respond with extreme hatred and avoidance. This seems like one of the responses that is reasonable ... as is acceptance and instrumentalization of the individual.
I agree a bio is a bad proxy, and I'm not saying I'm amazing but I'm not THAT incompetent. Most of my GitHub is pre-LLM era; github.com/lionkor
Edit: And like a lot of tools, HOW you use them is just about as important as the quality of the tool itself. Of course the tools can produce massive amounts of bad quality slop, they can also produce fast, focused edits that make sense.
Oh I know, I'm just trying to say that I agree fully that a bio is a terrible proxy, while also showing what I think is a good proxy for skill (open source work before LLMs became this "good", for example). I'm a big proponent of show, don't tell, and LLMs have kind of ruined this.
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.
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)
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.
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)
omp is not really pi. Its a very distant fork of it. And its philosophy is completely different, it includes everything you need except workflow skills. Personally, I think it is the best agent coding tool by far, and its what I settled on after trying 6+ different tools.
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.
Possibly depending on who's routing it, since Openrouter seems to have added a separate slug for this, and they don't have an alias router for the DS models yet, so it shouldn't have swapped over to it in case you use them for routing.
If you are directly using it through Deepseek, they do mention that the flash slug has migrated over
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.
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.
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.
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.
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.
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.
I should have worded it differently. Having a model selector as `string` seems like the wrong choice. At the very least, I should be able to select 4.X, like how we select dependencies, so people who rely on specific behavior can pin to 4.0 and I can say version >= 4 if I don't care.
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.
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
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)
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
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.
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..
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.
> 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.
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.
The culture is different, if you tried pulling any of the shit Sam and Dario do on a daily basis in China you'd get your wings clipped pretty quickly, much smarter to keep your head down as much as possible.
And maybe that's not a bad thing tbh because look at the state of the US right now
That's partly. I'd say the other part is it's an open secret they have "illegal" Nvidia GPUs and other "secrets". There's a common understanding not talk about these things because it'd hurt everyone collectively.
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.
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.
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.
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).
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.
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).
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.
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
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.
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.
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.)
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 :)
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
That approach won't give you access to the "Pro Mode" nor "Deep Research" though, which if you're paying for a Pro subscription already, "Pro Mode" is probably the top-top when it comes to "extensively researched answers & plans" across the entire ecosystem. That said, I do ask it to produce a .md plan that later is used with agents :)
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.
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
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 :)
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.
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.
I really like OpenCode - BUT: there is a loophole the size of Portugal in that ZDR language. All they say is that their providers follow a ZDR policy. I haven’t found anything promising that OpenCode themselves don’t retain Go usage data. Something to be mindful of.
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.
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.
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.
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.
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..
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.
What I and my team have been doing in papers is just to refer to the OpenRouter slugs. It upsets reviewers because they will complain it "is not sufficiently clear to a wider audience" but I do agree it's the cleanest approach.
Also goes to prove how much power OpenRouter has actually...
Not important, just a curiosity. I would expect that kind of knowledge would be pretty well burned into the weights, but what do I know about LLMs. Gemma 4 can answer this question well.
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.
I think it does not matter. Most people who use these types of models are into the IT world, and will know/be informed very fast that there is a difference. People will likely also use the 0731 behind it.
The only issue i see, is 3th party providers that have not yet updated. But that is going to be a short time periode. There is no reason to not update.
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.
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.
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.
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.
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.
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.
The memory bandwidth of the 2x RTX Pro 6000 Blackwell setup will be 10x higher, which should have an equivalent effect on the generated tokens per second.
Currently the 3bit (and 2 bit) quant on DGX spark (on one of them) and the M5 Max should just start. Right now. (I'm hoping to get an M5 Max delivered on monday, let's see if it happens this time. It's 2+ months since I ordered now)
The 4 bit quant technically fits (there's a 127 GB version) but ... obviously that's not going to work. It is so close though, surely someone will a way to do it.
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.
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).
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).
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..
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
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
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.
Ah, my bad. Yeah that makes sense. They do say "The official V4-Flash natively supports the Responses API format and is specifically adapted for Codex.", so at some point someone will make a "same harness" comparison.
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.
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)
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.
In terms of open research, China has absolutely overtaken the US.
Undermining out entire economy by subsidizing the release of DIY versions of our main economic drive sounds like a huge win for China.
Yes this is why I referred to 'months' and was downvoted by people who can't distinguish their politics from reality. You are restating exactly the text you are criticizing.
This is the nature of mechanical parrotlike repetition of propaganda:
a) You can have 'frontier' models with closed weights. Your sentence is basically a contraction within itself, again, the weight of ideology.
b) No one actually knows whether or not the closed weights of Bytedance are beyond all existing frontiers. Again the weight of ideology blinds you to the fact that the real 800 lb gorilla of Chinese Ai is more closed than Anthropic.
The parent made a totally coherent commment that China closing off their open weight models would cede the frontier (and majority of AI users worldwide) to the USA. Nothing you said counters that point.
In fact the overwhelming weight of AI use in China, the chatgpt so to say, is Bytedance's AI which is absolutely closed and uniquely opaque.
The press treatment of these matters was no good and they are slowly walking it back, e.g. NYT yesterday finally actually read the speech.
> We should put in place laws and regulations, technological monitoring, early warning and emergency response systems in order to strengthen the line of security, prevent abuses and malicious use, and ensure that AI is always under human control. In the meantime, we should jointly oppose overstretching the national security concept in the field of AI and placing one country’s security over that of others.
There was a blog post by a state-linked broadcaster cited in the NYT piece:
> “China supports openness, but this does not mean it advocates for the unconditional proliferation of all capabilities,” the blog said.
The article also quoted an American journalist:
> “If these models do reach those dangerous capabilities, they are not going to let it be a free-for-all in terms of releasing them," Mr. Sheehan said.
But a distinction has to be drawn between real dangers (which many people believe LLMs have not actually shown, to date) versus "dangers" hyped up marketing purposes or domestic regulatory-capture motives. Presumably the Chinese government is less interested in the latter.
It is just a question of being unaffected by the motives of the speakers, which is what adults learn to do.
But the speech doesn't say that? I'm looking at the full text. For example:
> We should take seriously the various types of inherent and secondary risks that AI may trigger.
The risk portion of the speech is focused more on application-level risks than model capabilities. Which is a much more sensible regulatory framing than what Silicon Valley has proposed.
The administration blather about banning open weights is characteristically confused. Xi has already stated (what is obvious) that he will ban security-endangering weights and keep them a state secret.
It's so good that the US government is rushing to ban all Chinese models as fast as they can.
The interest around K3 is a lot more defensible because we actually know what the architecture looks like, how much effort it takes to get it to run, and what kinds of results it gets on cyber evaluations. And no, it's nowhere near "dangerous" enough to where people might honestly want to ban it for real safety reasons. It does a good enough job at fixing cyber issues, but that's hardly a safety concern.
I’m curious to see why you’re so eager to ban open-source LLMs while being okay with Anthropic’s control. Do you think they’ll rebel against you?
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.
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.
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.
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?
I use OpenCode mostly (uses fewer tokens than Claude Code) and I am looking forward to the release of DeepSeek’s own coding harness.
It's replaced the Kimi models for me though.
- 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.
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.
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.
The main difference is that deepseek is bad at prose, and Codex models are much more eager to use tools provided to them (which is usually fine, they often make like 3 todos via tool calls for a simple one step task which is silly though), and deepseek in general benefits from good instructions more than GPT models maybe do.
Deepseek becomes much better if you give it tools for asking clarifying questions, doing self-review with subagents, and so on. The more tools it uses, the better the signal-to-noise ratio in its context, and the more consistent the output.
You must still review 100% of the code as if it's trying to sell you insurance.
Then you, dakolli - out of generosity and minute-to-minute devotion to enlightenment - sacrifice time from your busy day to sit down (though perhaps that's been painful lately?) or stand up with your phone - and offer a profound, deeply thought-out counterpoint in the following form (and I'll paraphrase):
"Nah mate, it's shit. All LLMs are shit."
Well, I guess having "Two PhDs, Three masters degrees. Expert in everything.." in your bio makes you very smart!
Edit: And like a lot of tools, HOW you use them is just about as important as the quality of the tool itself. Of course the tools can produce massive amounts of bad quality slop, they can also produce fast, focused edits that make sense.
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.
(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)
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)
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?
If you are directly using it through Deepseek, they do mention that the flash slug has migrated over
Crazy.
* 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.
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.
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.
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.
Does this make sense?
https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
I admire DeepSeek's openness, but even they have been raising prices after their discounts.
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..
I guess using a ZDR provider is good enough for now.
V4 flash cache read is $0.0028 per mtok
That's not "a bit cheaper", just saying
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.
And maybe that's not a bad thing tbh because look at the state of the US right now
That's partly. I'd say the other part is it's an open secret they have "illegal" Nvidia GPUs and other "secrets". There's a common understanding not talk about these things because it'd hurt everyone collectively.
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.
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.
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.
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.
At the speed Flash has on most providers, it doesn't really turn into a latency concern.
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.
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.)
No mention of weights, just API. When will the updated weights be released?
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.
> Trains: this provider may use prompts for training and may retain prompt data.
https://openrouter.ai/rankings?view=day#leaderboard-table
These days cost per task is more important, and SOTA models have become expensive.
These massive jumps in cheap models, is really great times!
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
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 :)
3rd party providers on OpenRouter can be cheaper but it's already so cheap.
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.
GPT 5.6 Luna is a new model in Go since I last checked, for example.
I still find this today:
> The plan is designed primarily for international users and provides stable global access. Your data will not be used for model training.
> because we added the new deepseek which we do not yet have a ZDR with we cannot blanket say we offer ZDR
I wonder how the website can make the statement that data will not be used for training.
Played around for a few hours and used up 80 cents of tokens.
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.
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.
Why not call it V4.1?
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..
The `-0731` style suffix is worse compared to a proper version bump like V4.1.
https://news.ycombinator.com/item?id=49082022#49087112
How is that important? Maybe it does, so what?
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.
The only issue i see, is 3th party providers that have not yet updated. But that is going to be a short time periode. There is no reason to not update.
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.
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.
You can also do 2x 6kPRO in a workstation, for ~20k.
But still, even for mid level projects API is orders of magnitude cheaper, since you don't need to set it up and maintain it.
The 4 bit quant technically fits (there's a 127 GB version) but ... obviously that's not going to work. It is so close though, surely someone will a way to do it.
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).
https://x.com/SourceCodeplz/status/2083099712760987746
i prefer GPT-5.6 Luna honestly
• 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!
Terra 87.4
https://openai.com/index/gpt-5-6/
> 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.
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.
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.
Wow that's crazy
Good times for those that don't need strict data protection