Thread

Grok 4.7

609 points · 541 comments · meetpateltech

meerita · 1790009756 · focus · HN ↗
Grok it's really expensive. I'm getting really amazing results using DeepSeek 4.1 Flash for fraction of the price.
1. testfrequency · 1790010532 · focus · HN ↗
  What is the most secure way to use this model as someone who is lazy
  1. user43928 · 1790011529 · focus · HN ↗
    I understand DeepSeek 4.1 Flash is available on US providers with Zero Data Retention if that is what you are asking.
    brcmthrowaway · 1790013281 · focus · HN ↗
    Link for the lazy?
    thehamkercat · 1790016196 · focus · HN ↗
    openrouter, "together" provider is fastest (165 t/s at the time of writing) and has ZDR and all
    <a href="https://openrouter.ai/deepseek/deepseek-v4.1-flash?endpoint=57e1cb3e-9762-4ee1-a67c-259bd3af24b7#providers" rel="nofollow">https://openrouter.ai/deepseek/deepseek-v4.1-flash?endpoint=...
    nicce · 1790019298 · focus · HN ↗
    Sadly, there is no way to tell if this is running with real weights or being heavily quantized.
    thehamkercat · 1790062205 · focus · HN ↗
    just checked, yes they don't seem to provide that information, most other providers are advertising "fp8" which is okay, but "together" doesn't, so they are likely using fp4/nvfp4
    riquito · 1790020064 · focus · HN ↗
    Eh, it's also one of the providers with "blind" vision models <a href="https://mmoustafa.com/blog/so-you-want-to-use-openrouter/" rel="nofollow">https://mmoustafa.com/blog/so-you-want-to-use-openrouter/
    thehamkercat · 1790063807 · focus · HN ↗
    update: coreweave/fp8 is at 191 tps, launched this morning, but really bad cache hit rate (~60%), coreweave is good for privacy but let's hope they improve cache

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.