Nex-N2.6 ships open FP8 weights (321B, GLM-5.3-Flash based), cuts reasoning loops 16%→1.6%
Physical_Hat4022 · reddit · 2026-11-11
A Reddit post announces Nex-N2.6, an update focused on getting stuck less—fewer reasoning loops and fewer calls to tools that don't exist in the environment.
- 321B parameters, based on GLM-5.3-Flash, with public FP8 weights on Hugging Face
- Team reports repetition dropping from 16% to 1.6% vs N2.5-Pro in an internal failure-replay test
- Reasoning defaults to max, so check that before comparing token usage
The author is curious how the improvement holds up in full coding sessions.
More from Models
- Leak: Google's upcoming model lineup includes Hydrogen, Krypton, Helix and a video model — koltregaskes · 2026-11-11
- Same prompt head-to-head: Fable 5.2 spends 3x time, 2x cost for auteur flair — iamfakhrealam · 2026-11-11
- Krafton open-sources Raon-1 nano, a 1.5B on-device LLM distilled from MoE — DimitrisPapail · 2026-11-11
- Do typos discomfort LLMs? Tokenization quirk sparks debate — menhguin · 2026-11-11
- MiniMax H3 hits 100 days open-source: 1,472 derivative models, 148M+ downloads — linoy_tsaban · 2026-11-11
- Meituan's LongCat-2.5 trends on Hugging Face as MoE multimodal model — meituan-longcat · 2026-11-11