BREAKING
Ornith-1.0 Released Under MIT
0
B
smallest
0
B
flagship MoE
0
models
RL Learns Its Own Scaffolds
1
Reason about task
↓
2
Generate scaffold
↓
3
Optimize solution
SWE-Bench Verified Scores
9B Dense
69.4
35B MoE
75.6
397B MoE
82.4
0
Terminal-Bench 2.1
0
SWE-Bench Pro
0
Multilingual
Fully Open Weights, MIT License
AI NEWS BLITZ
DeepReinforce has unveiled Ornith-1.0, a family of self-improving open-source coding models.