LisanBench creator @scaling01 says Claude Sonnet 5 is twice as expensive as GPT-5.5 and 57 times pricier than DeepSeek-V4-Pro

Price per token != cost per task

Oh my god, Sonnet 5 was MORE EXPENSIVE THAN FABLE to run the whole bench 💀

@theo It hasn’t been done voluntarily but they can un-launch a model

Sonnet 5 goes straight into the garbage bin

> 1.2x more expensive than Opus 4.8 Max > 2x more expensive than GPT-5.5-xhigh > 5x more expensive than GLM-5.2 > 7x more expensive than Kimi-K2.6 > 57x more expensive than DeepSeek-V4-Pro

How is this possible? Efficiency. It generates almost 2x as many tokens as Opus 4.8 per task.

stop scamming us with those shitty 500B yapper models

I WILL NOT SHUT UP AS LONG AS SONNET 5 IS MORE EXPENSIVE THAN OPUS WHILE DOING LESS

It gets worse

@mweinbach lmao

@f4micom omfg

@theo Only Anthropic would be able to achieve this. Launch a next generation, smaller, weaker, cheaper model, that is also more expensive.

At least we don't have fable 5 to compare it to, it would be funny if it was also cheaper then sonnet 5.

@theo At least we got something from Fable back even if it's just the price

difficult to understand

@theo This too. Gloating over 0 on CuberGym

@theo They gotta be doing this on purpose

@theo Is sonnet 5 just mythos in a trench coat?

@scaling01 all Anthropic models are very token thinking heavy compared to GPT.. check out http://matharena.ai too. Fable uses so much more than Gemini 3.1 & GPT-5.5

@theo Still working on some better graphs but basically we got a worse Opus 4.7 with this release...

@theo Seems like a weird release, not really understanding the logic

@theo But... you just told another user that Sonnet 5 was a better choice against GLM 5.2 due to it managing tokens better?

So, this is a take-backsies moment?

@mweinbach @theo ouch