Metis-1.6 post-trained reasoning and chat models (metis-chat/v1, <think> and direct modes, 131K context): Praxis (2.94B) and Logos (10.23B).
AI & ML interests
Open models trained from scratch: new architectures, data efficiency and fast local inference.
Recent Activity
View all activity
Organization Card
Lernex
Metis-1.6
Two open reasoning models that run every token through the same layers twice. Apache 2.0.
Pick the build for your machine
pip install dromos
dromos run praxisDromos is the inference engine written for Metis, for NVIDIA, Apple silicon and CPUs. dromos doctor shows what fits your machine.
MacApple silicon, MLX
Praxis 4-bit2.0 GB
Logos 4-bit6.6 GB
NVIDIA BlackwellRTX 50, B200 · NVFP4
Praxis NVFP42.0 GB
Logos NVFP46.6 GB
NVIDIA Ada, HopperRTX 40, H100 · FP8
Praxis FP83.5 GB
Logos FP811.2 GB
Full precisionBF16, the reference weights
Praxis BF166.2 GB
Logos BF1620.8 GB
llama.cppGGUF · support merging upstream
Praxis Q4_K_M2.0 GB
Logos Q4_K_M6.7 GB
Fine-tuningthe pretrained base models
Praxis Base6.2 GB
Logos Base20.8 GB
Serving with your own stack: pip install vllm-metis or pip install sglang-metis.
models 21
Lernex/Metis-1.6-Logos-GGUF
Text Generation • 10B • Updated • 136
Lernex/Metis-1.6-Praxis-GGUF
Text Generation • 3B • Updated • 125
Lernex/Metis-1.6-Logos-MLX-8bit
Text Generation • 10B • Updated • 23
Lernex/Metis-1.6-Logos-MLX-4bit
Text Generation • 10B • Updated • 24
Lernex/Metis-1.6-Praxis-MLX-8bit
Text Generation • 3B • Updated • 21
Lernex/Metis-1.6-Praxis-MLX-4bit
Text Generation • 3B • Updated • 22
Lernex/Metis-1.6-Praxis
Text Generation • 3B • Updated • 13
Lernex/Metis-1.6-Logos
Text Generation • 10B • Updated • 14
Lernex/Metis-1.6-Logos-NVFP4
Text Generation • 6B • Updated • 3
Lernex/Metis-1.6-Logos-FP8
Text Generation • 10B • Updated • 2
datasets 0
None public yet