Log inSign up
Log inSign up
Marcin Kliks
3,833 posts
Marcin Kliks profile banner
@vi4m

Marcin Kliks

@vi4m
AI Coding Principal Engineer at @allegrotech. I'm testing local AI and coding agents, Spec Driven Development daily (Qwen, DeepSeek, OpenCode, sovereign AI)
Poznan, Greater Poland
marcinkliks.pl
Joined July 2009
502 Following
755 Followers
RepliesRepliesRepostsRepostsMediaMedia

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone
or
Log in with username or email
Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
  • @vi4m
    Marcin Kliks
    @vi4m
    22 Sept
    Can’t wait to test it at Allegro scale tomorrow. Personally, Luna is my type for spec-driven development, where the specification is complete and non-negotiable. “She” behaves like an almost “free” supply of tokens—I don’t even think about the cost when plugging it into my
    @OpenAI
    OpenAI
    @OpenAI
    22 Sept
    Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe. GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale. We’ve also made caching and inference more efficient, and
    00:00
  • @vi4m
    Marcin Kliks
    @vi4m
    21 Sept
    Building in public — Code Explorer, day one. How do you solve cognitive debt for PR review? -> My first attempt: 3D visualisation of PR changesets. THIS IS THE WORST IDEA EVER.
    00:00
    1
  • @vi4m
    Marcin Kliks
    @vi4m
    21 Sept
    I stopped paying for software because of the security reasons 🛑 - I *generate* what I need with Astra model faster than reading license terms - If there is an open source soft repo I use it as “the inspiration” for codex - It’s safer than trusting vibe-coded repos with 2
  • @vi4m
    Marcin Kliks
    @vi4m
    20 Sept
    Remember Dflash2 for Qwen 3.8-27b? *Splash* is even faster - at least 60% faster on my mac (M4 Pro, 48GB) * and it's already integrated into LMStudio (beta) * so you don't need to switch to llama forks, or omlx all the time Qwen is winning this quarter and becomes de-facto
    00:00
    @zhijianliu
    Zhijian Liu
    Inco AI
    @zhijianliu
    19 Sept
    00:26
    Splash is here! A month ago, DFlash 2 ran Qwen3.8-27B at 70 tok/s on an M5 Max. Today, Splash runs it at 144 tok/s on the same Mac ⚡ Getting that speed used to take work. Splash is built around the model, with the DFlash 2 draft already in it. Two commands and you're running.
    1
  • @vi4m
    Marcin Kliks
    @vi4m
    19 Sept
    Breaking News: Claude Code now works with AGENTS.md! Great win for interoperability. No new for symlinks. Public pressure works for big corps!
    1
Edit with