Harness
opencode is strongest with qwen3.8-flash, 5th overall.
Valid evaluations covered 15 of 19 models. The other 4 have not been evaluated yet.
Most often 1st, in 12 tasks.
Each row is a kind of work and each dot is one pairing's place in it; further left is better. Hover a dot for its name and place, or pick a name below to follow one pairing.
The number in each cell is a place: where that pairing ranked in that task, 1 being best, so 3 means 3rd. Hover a number to see how many pairings were ranked there. A dash means the pairing was not run on that task.
1st2nd3rd4th5th6th and below