I Tested Sonnet 5.5 Against Opus 5.5. Is It Actually Cheaper?

Anthropic's Claude Sonnet 5.5 is supposed to be a strong, cheaper alternative to Opus 5.5, and one benchmark even puts it ahead in agentic coding. I ran it through my own animation challenges to see how that holds up. Several attempts used a lot of output tokens and stopped before producing a usable result. A tiny train animation looked promising once Gemini helped fix it, and the Cybertruck rally was genuinely good. I also tried a new "Neon Unicorn Frappuccino Challenge," but it failed once again, so I was stuck just admiring the GPT Sol 6 and Gemini 3.8 Flash results, which were both excellent... especially the Gemini 3.8 Flash one. By the end, I'm not convinced Sonnet 5.5 is the bargain it appears to be, especially given that I was paying for a lot of attempts that merely consumed a lot of tokens without producing results (under the 120K output token limit that I set in my tests) #claude #sonnet #anthropic Chapters 0:00 - Why test Claude Sonnet 5.5 against Opus 5.5? 1:17 - Animation attempts run out of output tokens (120K max allotment) 3:54 - Fixing the broken 2 KB train animation 6:10 - Cybertruck rally finally delivers 7:09 - "The Neon Unicorn Coffee Shop" challenge 8:48 - Gemini's delightful "Unicorn Catastrophe" animation 10:07 - Final verdict on cost and usefulness