3/ What early testers measured, as quoted by Anthropic.

Base44, 118 app builds: 3.6 iterations per build. Opus 5 took 7.7 - Balyasny, 2,441 finance tasks: about 121K tokens per answer. Sonnet 5 used 497K - Box: 2.4x faster than Sonnet 5, 12% fewer tokens - Zendesk: support

Parker RexSep 281

2/ Same price as Sonnet 5.

Less per task. - $2 per million input tokens, $10 output, $0.20 cache reads - Opus 5.5 is $4 and $20 - Anthropic: up to 30% less per task than Sonnet 5, because it uses fewer tokens - 30%+ faster output - At medium effort it beats Sonnet 5's best

Parker RexSep 281

1/ The scores, from Anthropic's own page.

Sonnet 5.5 against Opus 5.5. - Terminal coding: 70.6% to 66.4% - CursorBench: 55.5% to 57.8% - Computer use: 80.1% to 81.8% - Knowledge work, GDPval-AA: 1844 to 1846. GPT-6 Sol: 1487 - Reading charts: 61.6% to 64.4%. Sonnet 5 scored

Parker RexSep 283

Claude Sonnet 5.5 is out.

It costs the same as GPT-6 Sol and half of Opus 5.5. On Anthropic's terminal coding test it scores 70.6%. Opus 5.5 scores 66.4%. Sonnet 5 scored 10.3%. Thursday I called Sol the better deal. It has a challenger at the same price.

Parker RexSep 2811