Log inSign up
Cristian profile banner
user avatar

Cristian

@XCSme
๐Ÿ”ฅ uxwizz.com - Self-hosted Analytics with Session Recordings ๐Ÿฅ‡ aibenchy.com - AI Benchmarks 15+ years SWE experience. Love ๐Ÿ“
Amsterdam
uxwizz.com
Joined October 2016
198
Following
355
Followers
RepliesArticlesMedia
  • user avatar
    Cristian
    @XCSme
    Aug 15
    As good as Qwen 3.8 27B is, running it locally on a 3090 costs 3x more than GPT-5.6 Luna just in electricity costs... It is still cheaper than Terra, which it surpasses in intelligence.
    Image
  • user avatar
    Cristian
    @XCSme
    Aug 15
    I tested Qwen 3.8 27B locally on a RTX 3090 24GB. It works!! GPT-5.6 Terra level model running locally, no internet needed! It can code, use the browser, understand screenshots and images, parse documents, everything you need, locally!
    Image
    Image
    user avatar
    Qwen
    @Alibaba_Qwen
    Aug 14
    We promised open weights for Qwen3.8. Now, time to meet them! ๐ŸŽ‰ โšก Qwen3.8-27B: - A native multimodal dense model. With just 27B parameters, it outperforms Qwen3.7-Plus overall and shines in real-world coding & office workflows. - 262K native context, easily extendable to 1M