BREAKING
Qwythos-9B Adds 1M-Token Context
0tok
context length
0B
parameters
0M
CoT tokens trained
MMLU Jumps to 0.575
base Qwen3.50.232
Qwythos0.575
0pt
GSM8K strict gain
0/7
agent tasks passed
0k
GGUF likes
Strengths vs Trade-offs
Strengths
Apache 2.0 open weights
Native tool calling
Strong local coding
Trade-offs
Heavy VRAM at full 1M
256k-512k realistic on H100
Slight short-input quality dip
Small Open Weights Close the Gap
AI NEWS BLITZ
Empero AI has released an open 9B model that handles roughly one million tokens of context.