Conquered Arena: Just days after Anthropic’s Claude Opus 5.5 launched, it climbed to #1 on the Text Arena overall leaderboard—about 1509 points (±12), with preference points built up via human blind, one-on-one voting. In the top ten, Anthropic took up most of the seats.
This version came out on September 22. Input is about $4 per million tokens, output about $20, roughly 20% cheaper than the previous Opus 5. The context window is up to 1 million tokens. On the benchmarks, Terminal-Bench is about 66.4%, and CursorBench about 57.8%.
Arena also broke down writing style: dashes dropped from about 15.2 per thousand words to 0.8 (down about 95%); semicolons were reduced a lot as well, with sentences getting shorter—yet each single response is actually a bit more verbose (about 453→481 words). The leaderboard’s top spot changes quickly, so let’s make a note of it.
#AI #Anthropic #Claude
This version came out on September 22. Input is about $4 per million tokens, output about $20, roughly 20% cheaper than the previous Opus 5. The context window is up to 1 million tokens. On the benchmarks, Terminal-Bench is about 66.4%, and CursorBench about 57.8%.
Arena also broke down writing style: dashes dropped from about 15.2 per thousand words to 0.8 (down about 95%); semicolons were reduced a lot as well, with sentences getting shorter—yet each single response is actually a bit more verbose (about 453→481 words). The leaderboard’s top spot changes quickly, so let’s make a note of it.
#AI #Anthropic #Claude
