running Ornith-1.0-35B on a single RTX 3060. 12GB VRAM, 16GB RAM.
35B params. 170K context. ~52 tok/s.
been running it as the engine behind my Hermes agent and Qwen Code, and honestly it feels better than Qwen3.6-35B-A3B for that kind of agentic/coding work.
full llama-swap config:
llama-server
-m ornith-1.0-35b-Q4_K_M.gguf
-ngl 99 --n-cpu-moe 24 -c 170000 -fa on
-np 1
--cache-type-k q8_0 --cache-type-v q8_0
-b 2048 -ub 1024
--temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.0 --presence-penalty 0.0 --repeat-penalty 1.0
--reasoning on
@ornith_
Apocalyptic bird nest.
A Russian glide bomb knocks down a tree in Donbas. From the shattered branches rolls out a tiny bird’s nest.
Made of drone fiber-optic cable.
Source: Oleg Malchenko
Qwen3.5 has 3 models in the top 5 of Humanity's Last Exam.
- 27B dense: 48.5
- 397B-A17B MoE: 48.3
- 122B-A10B MoE: 47.5
Dense and MoE. 27B to 397B. These are perfect models for local inference and fine-tuning.