Inkling Mini - 26B A3B

As good or better than Qwen 3.8 27B on most benchmarks. Looping architecture - it loops a total of 3 times per token.

1 Like

I too am interested in seeing this optimized for 2xr9700s.