Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- If it matches GPT-5.4 on coding tasks (as benchmarks suggest) this could be my forever model. And with partial SSD streaming, I could run it locally today. :Dby Archit3ch
- The weights were just released a few minutes ago: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731by coder543
- the unsloth GGUF at <165GB will run on most 256GB RAM pure-CPU systems (or with llama-server and a mix of loading as much as you can onto a single 32GB, 48GB or 96GB GPU and the rest onto system DRAM).by walrus01
- Hmm, it's targeting a HW accelerator with a 128x128 matmul primitive, which one is that? Warp Group Matrix Multiply Accumulate on H100?by WithinReason
- Can't wait for the DwarfStar quants - I have been using DeepSeek v4 flash (preview) as my main coding agent for months now (running on my 128gb mbp) - it seems this model outperforms GLM 5.2 on nearly every metric. Thanks for sharing the news, I was refreshing huggingface but gave up thinking it likely would take some more time.by kamranjon
- For people with single RTX PRO 6000 96GB or DGX Spark 128GB, vllm-moet is a very good engine, although lesser known. It auto generate a symmetric 2-bit plane for inference and also generate a 4-bit delta cache to recover precision. Support ssd streaming oversized weight. You pick how much VRAM to allocate to each to balance out speed vs precision. 170 tps with ds-v4-flash demonstrated.
It use the stock model, no new models requires.
Worth spend a few hours to try.
The DGX Spark requires a small hack to ignore the difference between sm120 vs sm121, but it does run on sm121.
by ycui7 - It’s exciting that a model scoring this high is dirt cheap.
It’s also so inefficient, when they release the full performance numbers it’s not going to be good.
One example, it takes about 3.6x more tokens to finish the same work as Gemini Flash 3.6.
by WhitneyLand - Roughly equivalent to Gemini 3.6 Flash in capabilities at 1/20th the price...
Mind you, until the recent price cuts to Luna - Gemini 3.6 Flash wasn't even egregiously priced (but oh how things change in just 1 week).
- you are correct. but in my experience -- not benchmarks -- g flash 3.6 is SO BAD for coding. I'm using all vendors all day and gemini is the worst by far. I built my own semi-deterministic orchestrator for coding agents.by ptole_my
- > inefficient
That depends. Is it also more reliable?
If two books, one big one slim, prove the same thesis, what I would be interested in is the quality of the content, not the size. There can be a measure of efficiency in "have you really thought it through", but it is clearly complex - it requires measuring how solid the reasoning is.
by mdp2021 - Using tokens to evaluate models is an outdated approach. Cost per task is what matters. Not all tokens are created equalby Bnjoroge
- This seems to me like this is probably at least a large part of what OpenAI was up to yesterday with their aggressive price cutting; trying to get out in front of this.
If the full non-flash model follows up with the expected improvements, and at the price point they've been keeping, it puts the frontier labs in a tough position and it feels to me like like OpenAI is reaching deep into their pockets to try to head that off.
TFA link is a 404 though. I'm reading through https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731 instead
- So GLM 5.2/Gemini 3.6 level intelligence for $0.28/m output. And their updated Pro model coming soon....
Plus a size you can genuinely run at home: Unsloth lossless Q8 at 162GB.
by scosman - i am scared for the PRO model
maybe it is Fable level
- Q8 is ~ 151gbby segmondy
- I would like to see your "home"by luckydata
- The really interesting thing about this is how big of a jump was achieved with just extra fine-tuning here. No structural changes to the model, just more data, compute and time. It makes me pretty excited for the future of small models - DS v4 flash is a relatively small model when compared to the class it's competing with, so likely similar gains can be made applying quality data/training pipeline to other smaller models.by kamranjon
- If deepseek v4 flash is beating DeepSeek V4 Pro, can we expect new V4 Pro which is on par with Opus 5 in couple weeks (even better if it beats Opus)?by throwaw12
- They have said that the updated final version of V4 Pro will be published soon.by adrian_b
- Yes, they said that the updated V4 Pro will be published soon.by adrian_b
- We can expect a new v4 pro, this was a footnote in the v4 flash announcement earlier.by lionkor
- I’m salivating at the thought of this!
Deepseek v4 Pro prices with Opus 5 perf would be freaking unbelievable!!
This is probably a dream.
by websap - I’ve been using v4 flash for an app I’m building [1] and it’s amazing how cost effective and good it is coming from having always used gpt, opus and sonnet models.
It’s so cost effective I can offer a generous free tier since my goal isn’t to make money with it.
by jmathai - Already beat Luna on price/task, by about 2x:
https://artificialanalysis.ai/models/deepseek-v4-flash?intel...
by WithinReason - Came here to see how it performs against Luna. Would love to see how they compare on specific types of tasks.by akurilin
- If only they would let you opt out of training use, it might actually be a viable option.by ComputerGuru
- Maybe I'm reading that incorrectly, but it seems to me the cost is on the X-axis.
First, your direct comparison, Deepseek V4 Flash 0731 (max effort) $0.03 (rounded up) per task @ index 50.
OpenAI Luna:
* high effort $0.03 (rounded down) @ index 46
* xhigh effort $0.04 @ index 49
* max effort $0.07 @ index 51
So I would say a fair statement would be "OpenAI Luna between 2x and 3x the price of Deepseek Flash, what you get is 2 to 5 times faster inference"
The cheapest OpenAI model that beats it is OpenAI Luna (max effort) $0.07 @ index 51 (if you take the rounding out it summarizes to triple the price for similar performance), but still close to 3x faster.
And can SOMEONE please tell artificialanalysis that using dark blue for both Deepseek AND OpenAI is an especially unfortunate choice of colors, especially today?
by spwa4 - Somewhat relatedly, how do the economics for Huggingface work? They must be hosting petabytes of models and datasets by now. I have downloaded quite a few “just in case”, only to replace them with the later iteration months later.
Does the file hosting actually cost peanuts when you do it yourself and the cloud has shattered my understanding of what it actually costs to deliver so much data?
by 0cf8612b2e1e - Maybe download numbers are relatively lowby jvuygbbkuurx
- Bandwidth is really cheap when you run your own infraby lima