Reflection 70B turns two: the fake GPT-4o killer that was just Llama 3.1
jacek2023 · reddit · 2026-09-29
In September 2024, Reflection 70B appeared out of nowhere, billed as an open-source model that "destroyed" GPT-4o, hyped via YouTube.
Then people actually downloaded and tested it: it turned out to be essentially a Llama 3.1 rebrand. The community gradually unraveled the mystery, making it one of the most infamous scandals in local-LLM history.
The author marks the two-year anniversary as a reminder to stay skeptical of current LocalLLaMA hypes like jev, OpenClaw, and TurboQuant.
More from Fun
- Browser-to-Browser Multiplayer Works, Even Unreal Tournament '99 Exists — banteg · 2026-09-29
- Dev jokes that adding yourself to the docker group is basically installing a rootkit — haydendevs · 2026-09-29
- Every academic has two citation counts: the one they don't care about publicly and the one they know exactly — prof_kamilov · 2026-09-29
- Professor's take: Reviewer 2 isn't a person — it's a role we all eventually play — prof_kamilov · 2026-09-29
- Meme: the anxiety of typing 'git push' and handing things over to Claude — tibo_maker · 2026-09-29
- Three GPT models all drew a pelican 'riding' a bicycle by carrying it on its back — jade_jade_jade_jade · 2026-09-29