Opus 5.5 vs Sonnet 5.5 vs Astra vs 6.1 Sol (Mario Kart test) All on the same prompt Sonnet used over 2x as many tokens as Opus in the same amount of time and imo looks the best of all four

mario kart is a better benchmark than half the official ones

yeah it shows quite a few creative things at once, i give them a very open ended prompt and let them do their best so this is all out for each too

open-ended is the right call, the differences only show up when you don't steer it

Opus 5.5 extra winner here

everything factored in like mechanics and it working perfectly yeah, sonnet looks so clean though

Opus looks like the best balance here. Sonnet using 2x the tokens for a similar result is rough

it does visually look better but yeah it eats tokens

I want to rewatch the Mario movie after this post

😂

Would be great if u can tell cost wise too... Wasn't sonnet expensive?.. i think Opus might create that in less cost than sonnet bcs of the iteration that u have to do in it

my testing is on here with graphs:

👀👀thank you

Opus wins due to being more complete in the full project with things like cc better items and stuff like that but just visually Sonnet looks better

🤔I see .. Informative indeed

Opus is the best

what sonnetlooks even better than all

burns tokens like though

2x the tokens for the best looking result is a fair trade on a game demo. on a long agent run that gap adds up fast

sonnet cooked wtf

didn't expect sonnet to beat opus 5.5 looks natural

Opus was much more complete with mechanics and menus, but sonnet does look incredible

The split you made in the replies is the useful part: Opus built the more complete game, Sonnet built the better looking one. Do you score completeness against a fixed list (laps, items, CPU racers, collisions) before you judge the visuals?

Sonnet's panel is the one that feels fast. The glowing track edges and the flame trail behind the kart do more for the speed than anything in the other three

Both Claude models looks really good.

as always opus cooks

2x tokens for a better result changes the cost math on every task like this. cheap models stop being cheap when they have to ramble

I’m always so impressed at sonnet, am I the only one?

I'm still rooting for Astra, but damn, Sonnet 5.5 looks gorgeous in that preview 😂

Sonnet was incredible, an absolute token gremlin though

claude 모델들이 미감이 좋아보이네요

Tell me when you drop the benchmark for these models defending you from Nintendo lawsuit 😂

Price ?

I see two slop models here