Sabitlenmiş Tweet
you're better off thinking for yourself, and be wrong, than not thinking at all
English
botblastcap
14.3K posts
Running models at home is cool if you're a hobbyist
But it does not make any economic sense.
"But it's private!"
Well, so is Venice.ai/api
You can get an API key and tell your agent to use deepseek-v4-flash-0731 from Venice in 2 mins and save thousands of dollars and hours of hardware/software config time.
(and you can switch between any/all models as they come out on the same API key rather than trying to update hardware every week)
tae kim@firstadopter
Get the Nvidia DGX Sparks right now before they sell out. They will sell out soon with the latest DeepSeek V4 Flash drop. I guarantee it. Like an idiot, I already spent my AI budget on a 128GB Macbook Pro M5 Max. I should have bought two DGX Sparks instead. AI is developing so fast!
English
@ErikVoorhees @AskVenice if it's trivial for any normie to verify and hence be convinced that Venice is truly private, we will onboard the next billion users
my 2 cents
English
@jesseproudman @AskVenice @sabrinaesaquino hi, saw you liked my comment so i assumed it's decent feedback for you guys.
could we also have the reasoning efforts of each venice supported model displayed? atm i have to call the API then parse individually to get the data. thanks!
English
works smooth now, thanks jesse
could try something like this (image from openrouter).
been getting a lot of server overloaded errors on not just deepseek v4 flash 0731 but also xiaomi mimo v2.5 and minimax m3
would be really helpful knowing where went wrong i.e. Venice's servers or our own setups e.g. at first i thought it was hermes agent that was buggy, then realized it was Venice's end.
perhaps also consider observability tools for better DX/UX. this is invaluable for most users that dig into logs from their agent harnesses running on Venice APIs.
English
@botblastcap @AskVenice Working on both additional capacity and a better public dashboard for when we’re hitting our limits.
Give it a whirl now.
English
@jesseproudman @AskVenice servers look overloaded for deepseek-v4-flash-0731
is there a venue we can check latency and throughput atm?
English
@jesseproudman @AskVenice could also consider defaulting the app's agentic chat to this model, free for pro users!
English
@jesseproudman @AskVenice chad
imo this will be magnitudes more significant than k3/glm5.2 given its cost at similar intelligence
English
@botblastcap @AskVenice Teams experimenting with it right now.
English
$0.28 out
…
I can’t imagine the load on deepseek’s GPUs right now
🔥🔥🔥
English
Official DeepSeek v4 Flash weights are out in @huggingface 🔥🔥🔥
huggingface.co/deepseek-ai/De…
English
now deepseek v4 flash can handle both daily use and complex stuff
qilua@qilua02
deepseek-v4-flash is honestly enough for my daily use, while glm-5.2 handles my complex stuff. Plus, they’re 3x to 10x cheaper than US models with the same level of intelligence. Just leaving my note here as of July 2026.
English
@imbabybrooklyn it takes 20s for a simply "hi" for me tho... bloated context?
does this happen to u?
English
The older I get the more I realize:
• Silence is a response
• Being unbothered is a skill
• No is a full sentence
• Most "urgent" things can wait
• Respect > Attention
• Apologies are empty without change
• Movement is medicine
• You're not for everyone, and that's fine
English
hey @ErikVoorhees @jesseproudman, how do users know if @AskVenice 's e2ee deepseek v4 flash is updated to use this update?
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex
DeepSeek-V4-Flash is updated. «DSBench-FullStack is an internal full-stack development test set, and DSBench-Hard is an internal Coding Agent hard-problem test set» «The official release of DeepSeek-V4-Pro will follow soon» These scores are… quite a lot better than Preview.
English