DeepSeek costs OpenCode Go user $1.14/day; dual DGX breaks even in 24 years
twitter.com
twitter.com
their recent changes to the privacy policy broke their promise of zero data retention. specifically, they offered chatgpt luna under a zero data retention privacy policy. luna was later shown to be 30 days retention.
their privacy policy has never guaranteed your prompts will not be logged and when asked they have failed to revise it.
when challenged about sending data to openrouter without listing it as a 3rd party processor they offered a dismissive response. the same with running prompts through cloudflare. seems trivial, but signifies general disinterest in security.
by default if you run the harness outside your config file by accident, it will automatically run silently with a free model that sends your prompts and local data to an endpoint with training enabled. on top of that it used to dump all the prompts sent to free models into an s3 bucket, the feature was literally called 'datadumper' in the source.
For instance, this "marketing" claim that it'll take 24y to break even if a user only uses 100m tokens/day (~$1.14 in DeepSeek v4 Flash usage) ignores the fact that OpenCode Go has 5h & weekly throttles. Besides, folks who self-host models usually run automated jobs [0]. I think the GPU setup could possibly serve 10+ "users" concurrently, bringing down the break even by 22y (10x).
[0] For comparision, we routinely do 200m to 500m tokens ($2 to $5) on merely 3 to 8 automated code reviews per day with DeepSeek v4 Flash on max.
> With Go, you pay $10/month and we aim to give you 6x that in usage.
For most models, we make this work through bulk discounts and reserved GPU capacity. We then pass those savings on to you through the 6x multiplier.
https://opencode.ai/docs/go/#why-some-models-have-lower-usag...
I think it's back to 2X usage, meaning it's cheaper token-burn than usual to use. Which is lovely.
OpenCode Go has been so nice to have. I love having access to DeepSeek, Qwen and MiniMax M3 when doing design work, to see what different models cook up. I've been very surprised with MiniMax M3, not as a particularly good architect, but at it's very good ability to state the problem elegantly & to frame the different decision points very well. That's been a fun ongoing surprise.
Maybe hallucination is good for prototyping and creative work, but maintaining and debugging code might be better done by a boring model?
Models are all non deterministic and local is the only gauntee that your investment can continue to pay.
Cloud models will continually change nondeterminism ontop of the model.
For example, I needed to write an invitation letter for immigration control for a relative visiting me. Previously I would have used a search engine for a template. Today I fire up my local qwen 3.5-9b for this kind of stuff and feed it all the private data I need.
Unfortunately it is unlikely the average user will known how to avoid this data collection. Even if the LLM is local you are likely feeding the prompts to remote servers if you harness/chat-interface is not properly vetted.
I eventually switched to LM studio and the same model runs much better, like 70tk/s.
Not sure if it was because I was running llama.cpp inside podman or badly tuned LLM arguments. But LM studio is unfortunately much more practical.
Although I agree with you. I do not really know what kind of telemetry LM studio is running and I would rather not be using it.
It isn't even really worth the (minimal) ops to stand up rented MI300Xs to sell excess capacity to them even if it was minimally profitable, when I tried I was content to give API keys to friends to beat on it.
Other providers don't.
You can also choose to route to ZDR only.
https://openrouter.ai/deepseek/deepseek-v4-flash-0731#provid...
Does this mirror what other people have been experiencing in waves?
I get 140 tok/s on short prompts on an rtx3090 on the qwen3.6 moe which makes is easily 4x the speed of Chatgpt or Claude doing Web research.
But it is a much simpler model. It is only good for simple queries, usually I search for cheapest product in stock in my country available online and stuff like that.
I use the dense model for planning and such, but on its own it is much inferior to for example opus. It needs careful pipelines that check facts and such and in such harness it can be used for mamy tasks.
I prefer to stay with my 3090s.
Clearly nVidia and others are gatekeeping technology from the pleb so that the rich who own the datacentres can charge us massive margins.
Oh the debt or not making an even they are supposedly "suffering from" is just a classic mechanism to avoid paying taxes.
I have a stack of ten or so 3090s sitting in boxes, but it's not worth the hassle to use them. You can easily run models as cheap as water in the cloud.
Sitting around 15 minutes for local Minimax is stupid when you're trying to be productive. You can spin up parallel job instances and multitask in the cloud.
If you want freedom, build open source cloud infra.
You rent your ISP line. Why isn't renting GPU compute seen the same way? You still have compete ownership over your stack, you're just letting someone else deal with the capital outlay and headache.
If you're just vibe coding with giant frontier models, yes, the value will be worse. Especially now, where GPU prices have spiked another 20% last month.
For some tasks where owning the setup and full kv cache matters, the payoff calculation is ridiculously in favor of running your own deployment.
For instance for some batch classifications jobs where the prefix cache hit rate will be >95%.
The calculus also changes if you just use AI as a light tool while coding and don't need the giant models; qwen3 27B runs at 80TPS on a 5090 properly deployed.
One of my tasks has a very large but very static system prompt and instructions, think something like 128k tokens. On our own hardware, we keep that cached, sprinkle in the 8k of goodness needed for output, and DSV4 Flash 0731 can generate our output at ~50 tps on $10k worth of hardware.
I've been benchmarking vs publics clouds (which are tbf insanely cheap!) and it's basically a break even in 24-36 months if nothing changes, which it could for better or worse.
Combined with the security and stability of internal hardware, to me it's in "no brainer" territory for this workload, even if models don't improve.
Our only "risk" is better cheaper hardware or cloud costs, which given the trajectory is murky at best short of a bust.
I mean obviously it's worth it just so you can flex on HN. But curious whether there was any other reason? Retired scalper?
We got destroyed by bigger players (ElevenLabs, Suno, and more recently Fish) that fundraised and had better in-house models.
It was a headache to run, I wanted to pursue a different market (video/creative), and I pivoted.
My pivot failed and was dead for over a year as I went too far into controllability. One of my changes then got a semblance of PMF and it's doing $4M annualized run rate (but it's not fundable in its current shape).
I ultimately want to build open source cloud infra and open source workflow products - more friendly than Comfy, more free than RunPod.
In any case, I can't run my current work on the GPUs. But I'm close enough to pretraining and fine tuning smaller models that I want to use them for future experiments. Not that I wouldn't do that in the cloud.
To be honest I doubt you'll ever use them again, especially if you have dismantled the cluster and put them back in their boxes again. The cluster would be cheaper than cloud but you don't seem that cost constrained.
Is it mostly seen the same way. What is unacceptable is removing the freedom (which you mentioned) of people who prefer to run models locally.
Just as people have the right to tinker at home with DIY and electronics, fully knowing they won't compete with the latest ASML machine, people are free to use open-weight models at home.
BTW Minimax H3 running local can generate amazing short vids quite fast: about 50 seconds to generate a 7 seconds vids on a 4090 (depending on the settings). I've got a friend who spams my Telegram daily with such (no censorship and NSFW btw) vids.
I don't run local but I'll defend the rights of people who want the freedom to do so and I won't look down at them from my high-horse talking about "electricity" and "productivity".