秋田预测市场 · Discord 审计

AI模型Claude与GPT 5.5性能对比引发技术讨论

2026-06-23 · Polymarket · 51 条相关讨论

blunkoed 2026-06-23 00:04:58

but tbh now i don't bother with juggling with 10 models. i just use gpt 5.5 high and never run out of usage on the pro plan

but tbh now i don't bother with juggling with 10 models. i just use gpt 5.5 high and never run out of usage on the pro plan

blunkoed 2026-06-23 00:05:14

it's much less sloppy than claude

it's much less sloppy than claude

gzgz 2026-06-23 00:07:20

literally i hit limit on claude daily

literally i hit limit on claude daily

gzgz 2026-06-23 00:08:52

pro

pro

Abhi 2026-06-23 00:09:29

oh yeah its not visible on UI

oh yeah its not visible on UI

blunkoed 2026-06-23 00:09:48

jeez wtf r u doing to reach limits daily

jeez wtf r u doing to reach limits daily

Coolnyul 2026-06-23 00:33:23

i have only been using claude, i was ok with opus 4.8, but after trying fable, it feels like a retarded version...

i have only been using claude, i was ok with opus 4.8, but after trying fable, it feels like a retarded version...

Coolnyul 2026-06-23 00:33:43

how much better is gpt 5.5?

how much better is gpt 5.5?

blunkoed 2026-06-23 00:34:35

Well it’s kinda impossible to quantify but in my experience muchhhhh better than 4.8 and codex cli is less bloated and shitty than Claude code

Well it’s kinda impossible to quantify but in my experience muchhhhh better than 4.8 and codex cli is less bloated and shitty than Claude code

Speculate 2026-06-23 00:50:37

imo claude smokes gpt anyday

imo claude smokes gpt anyday

gzgz 2026-06-23 00:57:55

for now i believe gpt 5.5 is better now by far claude allocated most of it's infra and resources to fable so opus is less performant ( that's why it's retarded , it has less resources than it had before )

for now i believe gpt 5.5 is better now by far claude allocated most of it's infra and resources to fable so opus is less performant ( that's why it's retarded , it has less resources than it had before )

gzgz 2026-06-23 00:59:58

AI models often appear less performant over time due to cost-cutting in production, "stealth updates" to safety or behavior guidelines, and shifting user expectations. When new, more powerful models are released, developers may alter older models to save on computing power, or simply re-calibrate their standards because they are comparing old models against newer, more capable ones. https://www.scientificamerican.com/article/yes-ai-models-can-get-worse-over-time/

AI models often appear less performant over time due to cost-cutting in production, "stealth updates" to safety or behavior guidelines, and shifting user expectations. When new, more powerful models are released, developers may alter older models to save on computing power, or simply re-calibrate their standards because they are comparing old models against newer, more capable ones. https://www.scientificamerican.com/article/yes-ai-models-can-get-worse-over-time/

blunkoed 2026-06-23 01:01:43

the main issue when anthropic is that they use 3 types of gpu to serve their models and they just can't do it properly

the main issue when anthropic is that they use 3 types of gpu to serve their models and they just can't do it properly

blunkoed 2026-06-23 01:01:54

when you use 4.8 on bedrock it's MUCH better

when you use 4.8 on bedrock it's MUCH better

Coolnyul 2026-06-23 01:34:08

Oh I see, Bedrock is an exclusive aws cluster, so Claude only runs on those servers and they supposedly have only one type of gpu?

Oh I see, Bedrock is an exclusive aws cluster, so Claude only runs on those servers and they supposedly have only one type of gpu?

Coolnyul 2026-06-23 01:35:19

Hmm, according to ai they also have multiple types of gpus

Hmm, according to ai they also have multiple types of gpus

Coolnyul 2026-06-23 01:35:34

So it should not be better then

So it should not be better then

blunkoed 2026-06-23 01:43:27

Yeah but I'm pretty sure 1. they have a much more robust system 2. you do not jump between gpu types mid session (my experience)

Yeah but I'm pretty sure 1. they have a much more robust system 2. you do not jump between gpu types mid session (my experience)

blunkoed 2026-06-23 01:44:01

we did pretty extensive testing for smth unrelated to poly and the difference was clear

we did pretty extensive testing for smth unrelated to poly and the difference was clear