OrcaRouter's same-day Qwen3.8-Flash-Next hit piece called out as AI slop on Reddit
buttplugs4life4me · reddit · 2026-10-13
A Reddit user flagged a blog post from inference router OrcaRouter comparing Qwen3.8-Flash-Next against Qwen3.8-27B, calling it AI slop: since the new model launched with no independent benchmarks yet, the post's entire argument amounts to "it's bigger, so don't use it."
The post's spec table shows Flash-Next is a 125B MoE (6B active per token, plus 51B N-gram) needing 96GB RAM, while the 27B dense model is Apache 2.0 and fits a 24GB card; the footnote admits 27B has independent arena results while Flash-Next has none.
The poster questions why a company published this on its official site the same day the model released, and adds that OrcaRouter was the only one of ten Qwen3.8 setups that failed in their environment.
More from Fun
- A24's 'The Bird' documentary by Nathan Fielder blends real and AI-generated footage — AIandDesign · 2026-10-13
- Unicorn prompt showdown: two models, two wildly different art styles — imjustnewatai · 2026-10-13
- Asking ChatGPT to show if Meta Wonder is good yields an instant 'Yes. Very.' — rose_e_wang · 2026-10-13
- Palantir CTO: some AI researchers now smoke, betting AI will cure cancer — moonsandhues · 2026-10-13
- Daily Codex user rips Codex Cloud as 'completely useless' after Grok one-shots the same task — zeeg · 2026-10-13
- Anesthesiology as intuition without rigor: a funny math/physics debate analogy — QuintinPope5 · 2026-10-13