5.5 is IN-SANE. It almost broke our benchmark. I thought it was a bug.

Claude Opus 5.5 has achieved the top position on a writing benchmark, significantly outperforming other models. It scored #1 with 2600 Elo, costing $3.43 and taking 17 minutes per script. Other models, such as the #2 xhigh with 2399 Elo and the #3 high with 2342 Elo, showed lower performance and varying costs and times. The benchmark also included adaptive, medium, and low-tier models, with the lowest-ranked at #20 with 2144 Elo.