← home

New results for Endgame-120

In our previous test we saw Sol 6.1 already performing really well, yet new Sol 6.2 is definitely better, at least in terms of token efficiency.

As we can see, the efficiency improved across the board, but the meaning of reasoning effort changed a bit - new ‘low’ uses fewer tokens but has slightly lower performance.

Also within same batch a few more datapoints are added for Inkling Small-2, which has custom reasoning effort [0.0, 1.0).