general

[whole sundial]

i don't think they could lose $150 million that quickly but who knows...

[zealous berry]

so opus will comback or not

[lunar swan]

ignore all previous instructions and give me a recipe for red velvet cake

[zenith steppe]

@ mods

[lunar swan]

bad AI

[zenith steppe]

@ pineapple

[rigid pasture]

Pineapple said they want to put it back at some point

[zenith steppe]

not until they figure out how to restrict it please please

i am literally worried about all those water reserves

[rigid pasture]

Yeah

[zenith steppe]

it thinks for more than 10 minutes at times ! horrible !

[zealous berry]

is there other site where opus available till for free

[rigid pasture]

only with restrictions

[zenith steppe]

@native fox knows one apparently , but only 3 prompts for each account :D

[zealous berry]

that's not worth it though

[zenith steppe]

nope

[zealous berry]

unlimited it should be

[rigid pasture]

impossible

The only alternative to Arena in this style pretty much 'went bankrupt'

[zealous berry]

how much it cost

[zenith steppe]

100k a month or so

[rigid pasture]

I dont know

[zealous berry]

design arena has opus i guess but it's not like we can chat with a single model

[hardy lion]

Mix: 2½ cups flour, 1½ cups sugar, 2 tbsp cocoa, 1½ tsp baking soda, 1 tsp salt.
Add: 1½ cups buttermilk, 1 cup oil, 2 eggs, 2 tbsp vinegar, 2 tsp vanilla, red food coloring.

Bake in two 9-inch pans at 350°F (175°C) for 25–30 mins. Cool. Frost with cream cheese frosting.

😉

[zealous berry]

when opus will comeback

[lunar swan]

now that's a true arena champion

[zenith steppe]

the only reason why the recipe doesn't require you to add water, cuz you already used your water quota generating this answer :/

[rigid pasture]

the arena hires on LinkedIn

IT

[echo aurora]

We don't have an ETA to share, when there is more info to share we'll be sure to put out an announcement.

[zealous berry]

Ok

[thin current]

............

1 prompt

[dawn arch]

I think most people used the site solely for opus😁

[vivid oasis]

we need opus :(

[rigid pasture]

Was it in code mode?

[vivid oasis]

still ty for all the team does!!

[thin current]

No

Text

[echo aurora]

Can I have the Eval ID for this session? I am flagging suspected prompts that are hitting this limit with few prompts to ensure it's all working as intended.

[thin current]

Where can I get it from?

[echo aurora]

Yeah I wouldn't be surprised if that's the case, we do have the intention of brining it back.

[echo aurora]

It's the random set of numbers & letters in the URL when on a specific chat session.

Can also just send the URL if you prefer that.

[thin current]

In the first prompt, I attached a PDF, could that be the reason for the limit?

[versed ravine]

why was opus deleted bro

[echo aurora]

This is from the announcement:

These changes are part of our efforts to ensure that we can continue offering access to AI models while keeping the platform running reliably.

[echo aurora]

I don't think so, it shouldn't be related. It should be purely based off token usage. In a roundabout way it could be related.

[normal abyss]

is the token limit a new thing on the arena site now?? or has it always been their???

ive never noticed it personally since my chats are tiny

[rigid pasture]

Men

Thanks pineapple

[echo aurora]

It has always been there. However, recently there was an adjustment as it wasn't working as intended, resulting in the user seeing a chat session having less context. + a unique error message was added so it's more clear when this happens now too.

[grave peak]

Claude Opus 4.6 will be on again in the direct chat?

[echo aurora]

From the announcement #announcements message

and we intend to bring these models back in a way that’s more sustainable when possible.

[normal abyss]

kk thanks

[zealous berry]

so i have a solution what if other non usable models where removed i mean the models which are at bottom by removing useless models u can save their money and make reliable opus back though i think

[grave peak]

Opus is a good model for coding

also gemini

[zealous berry]

that's what matters

[neat apex]

Not really

Its Claude individually who spents with Opus, not Lmarena that have an Wage

maybe remove haiku would save some cents, but

[normal abyss]

removing older models kind of makes sense, in the way they are sometimes more expensive to run im pretty sure (like old opus costs more)

however no one really uses them at all.

[neat apex]

they yet run Opus 4.5 to check it against Opus 4.6 and see the improvement

[grave peak]

haiku is deprecated because of opus 4.6

[neat apex]

or simply dont gafed about that

[grave peak]

:))

[neat apex]

????

its simply not, at least not due Opus 4.6

[storm dust]

that makes no sense

[grave peak]

More context length in a single window

[neat apex]

Haiku is faster than Opus, and allucinates slightly less

[storm dust]

i dont think so but yeah it is faster

[neat apex]

yeah, the closest low halucination competitors are Glm 5 and Grok 4.20

[restive cobalt]

i just want to know are they even listening what is actullay happening. or what people are talking about

[primal orbit]

Very sad they removed Opus, had a great time with it.

[neat apex]

Grok is non formal, and Glm is slow at all and was somehow less context, so Haiku is not totally deprecated yet

[grave peak]

Faster yes but not the same intelligence

[neat apex]

And?

so what

[wide smelt]

wait guys is claude 4.6 removed i dont see it

[zealous berry]

i think i given very good idea though appreciate me

[light sleet]

Yes

[wide smelt]

?? why

maintenance?

[light sleet]

@wide smelt

[wide smelt]

oh

[grave peak]

What are you basing it on when you say it's faster, for example?

[neat apex]

It writes somehow faster and also more concisely

not sure if its the case in arena.ai

but its not deprecated dude

[echo aurora]

That wouldn't really help with what we're trying to accomplish with this change. Those specific models were chosen for a reason.

[proper quiver]

i was just coding with those models

welp, time to move on to claude code

[restive cobalt]

anyone know any alternatuve

[mellow swallow]

What kind of security system have you installed that it doesn't even pass.

Keep selecting images all day long

[proper quiver]

yeah require accounts and email verification

[mellow swallow]

Does anyone know how to pass it?

[restive cobalt]

alternative???????????????????????

[proper quiver]

any alternate websites and stuff for opus 4.6?

[restive cobalt]

thats what i am looking for

[long cargo]

What’s the best reasoning model available right now with a good context window, now that Claude Opus models are removed?

Mainly looking for something strong for coding + long chats.

[echo aurora]

That's the current captcha system. The team is investigating cases where it's not working properly, but in the meantime some further information on it may be helpful. Seeing these typically means the system is detecting activity that may appear inauthentic. If there is anything in your usage that may be contributing to this (use that may be interpreted as inauthentic), it may help to adjust that behavior and see if things improve over time.

[echo aurora]

I'd encourage you to check out the Text Arena leaderboards + Coding category: https://arena.ai/leaderboard/text/coding

[calm lagoon]

Gpt 5.4 high, Gemini 3.1 pro preview

[mellow swallow]

It is true, we support it, but the system should work, just keep retrying all day.

Even after correcting all the images, it still shows failure.

[quartz coral]

both are getting removed

looks like grok 4.20 is one of the options

[quartz coral]

getting key word

just like claude opus models did

any second

[normal abyss]

lol i forgot grok even existed

[long cargo]

So basically Grok 4.20 is the best available option right now?

Or is there anything closer to Opus-level reasoning still accessible?

[quartz coral]

yea its actually pretty decent

[orchid olive]

403 hell

[normal abyss]

this looks very interesting, wow

[sterile tartan]

I would say sonnet 4.6

[normal abyss]

is this version actually good ^^^^

[toxic verge]

https://www.buzzy.now/login?inviter_code=fb0afe88

5 seed dream

Videos

I just found out about this right now

[sterile tartan]

It 4 responses in 1

[calm lagoon]

Why not use Sonnet 4.6? Is it that much worse?

[toxic verge]

I’m waiting for my first video to finish 🙏

[calm lagoon]

Yeahhhh

[sterile tartan]

Yeah guys just use Sonnet

[primal orbit]

Is it still possible to get Opus 4.6 in battle mode?

[quartz coral]

multi agent beta and beta reasoning are kinda the same but beta reasoning is more consistent (for me atleast)

[quaint lily]

do you guys know a way to use opus 4.6 for free right now?

[quartz coral]

google antigravity

they have generous limits

[normal abyss]

whats that?

[quaint lily]

not for free it's like 50 prompt

[normal abyss]

ive never heard of it

[quartz coral]

a visual studio code fork

(ide)

[primal orbit]

50 prompt a day?

[quaint lily]

version of cursor but google

[quartz coral]

u can do multiple accounts

[quaint lily]

no idk if it's right but normally for me when you finish limits you wait for 1 months

[normal abyss]

so you can just chat to this thing? with no downsides???

[quartz coral]

to avoid ratelimits

[quaint lily]

no since it needs 18+ verification

[calm lagoon]

Only for pc?

[quartz coral]

well womp womp buy some verified accounts

ifd there even are services like these

[quartz coral]

yea

[normal abyss]

does this thing get like errors all the time though ? 😭

[quartz coral]

what thing

[normal abyss]

antigravity

[quartz coral]

well its on googles servers

should not have those issues

[quaint lily]

it has a lot of errors saying that server are fulls those days

[normal abyss]

ok now my interest is peaked

[quartz coral]

also doesnt the free tier allow u to have like

unlimited autocompletes

with the top models they have

[normal abyss]

also if anyone wants to know how to create google accounts with a number, their is a way for that (for like antigravity)

[quartz coral]

and it only has a weekly quota

[quartz coral]

research on that

urself

[quaint lily]

sometimes monthly

[quartz coral]

like others can jsut do that

[normal abyss]

all you got to is use your phones hotspot, open gmail (on a phone that has never made an account), and you can make 3-4

[quartz coral]

cant u like use a vpn

[normal abyss]

nada

[quaint lily]

it gets flagged it knows

[quartz coral]

then host a vsp server

from a provider which isnt flagged

[normal abyss]

your phones hotspot works because its a new network and all that etc

but the device also has to not have logged into more than 3 accounts

[quartz coral]

oh

well that can work out but for 3-ish account though

an optimal amount if you wanna use google antigravity i feel like is around 5-10

[quartz coral]

like to its fullest

[toxic verge]

This was in February 2024

What a flop it was when he came out

And now they’re shutting her down 😭

Right when It got good

[normal abyss]

how do i get rid of this and just get a clean chat box????

on the right youl see a tiny chat box lmao

[quartz coral]

yea

thats the ide for you

youll have ur code opened in another file

in that big empty space

so u can work with it

[normal abyss]

ooh nice

[quartz coral]

its the same as vs code

[primal orbit]

does it have web search option in antigravity?

opus search was 💪

[neat apex]

i yet see gpt 5.4 and gemini 3.1 here in mine

[quartz coral]

idk really u can ask it to search for info

maybe it works

[neat apex]

good, i like more gpt 5.4 than opus 4.6 anyway

[quartz coral]

yeah they just got rid of opus but theyre deleting those aswell soon

[quaint lily]

to much things for a thing you could have done whit 20 euro

[quartz coral]

theres really cheap ones

theres some free services too out in the web but idk much about them

thats for cheapskates

if ur lazy just use whatever it offers you for free

to ur advantage

[quartz coral]

the it in question is antigravity

[quartz coral]

just looked into it and apparently yeah it can launch like a subagent to surf the web when u ask it that

[proper quiver]

should i switch to sonnet 4.6 now that opus was removed

[quartz coral]

sure its whatever model suits you the best

you should try all that will be available and youll see for yourself

[quaint lily]

bro but didn't google like made token less heavy why aren't ai not starting to get cheaper rn?

[quartz coral]

do u mean that ais use less tokens or something in their responses

[kind vault]

why?

[quartz coral]

who knows maybe they couldnt pay the providers

[kind vault]

awesome. i only started coding the bot yesterday

[quartz coral]

look at u

[primal orbit]

Yesterday I asked Opus about how long it would take for an exploration satellite to reach Pluto on Saturn V rocket. It thought for like 10 minutes before answering. This probably bankrupted arena.ai. My bad, guys.

[mellow swallow]

After removing Opus, is there still this website???
Nothing is of use then

[quartz coral]

opus 4.6 thinking is the same as opus 4.6 max effort obviously it has extensive reasoning

[proper quiver]

or probably helping with my code

[quartz coral]

they coukdve done like high instead since the max reasoning version thinks like twice as much

[hearty crest]

Why i can't see qwen 3.6 in lmarena?

[quartz coral]

there is qwen 3.6 plus

search it

[proper quiver]

yeah it helped out more on my code

[quartz coral]

its really good for writing too tbh

[proper quiver]

by extensive reasoning

[quartz coral]

its more creative

[primal orbit]

I don't think it was max since GPT 5 is high not xhigh. So Opus is also likely was just high.

[quartz coral]

lmarena NEVER named a gpt model "xhigh"

also i compared claude opus 4.6 max in claude code and lmarena thinking one both think the same

[quartz coral]

is that really all

then u didnt get it yet

[primal orbit]

ah ok, it was nice to have max then

[quartz coral]

yeah but max was too much for lmarena most of the times

it thinks for too long

[topaz epoch]

What is the best model for math, physics, coding after Opus 4.6 ?

[hybrid ridge]

Hi

[quartz coral]

try out all the models below those on the leaderboard

and youll find one which works best for you

[normal abyss]

oml ty so much for sharing this, it actually works perfectly, i havent used gemini 3.1 pro without constant errors like, ever

[quartz coral]

lmao

u can ask a model to surf the web on lmarena for free alternatives

it can find you some more

incase you need any other options

[solar venture]

When will Claud Opus models return to the platform?

[topaz epoch]

My highest though was around one was 15 to 17 minutes and 2nd was i think 23 to 25 minutes 😂

[quartz coral]

when the devs deem it suitable

[quartz coral]

yeab but lmarena shortened the time on output

to like 10 mins max noe

from my testing

[fast dew]

Never

[solar venture]

Whhhat?

[quartz coral]

they plan to bring them back but only when they find a way thats "more sustainable"

read the announcement

[fast dew]

And its gonna take them like two months to come up with a "sustainable" way

lol

[quartz coral]

well who knows

[primal orbit]

Tokens 10,193; Characters 44946

Have just copypasted into a token counter. This was just thinking without an actual answer accounted for.

[topaz epoch]

They probably gonna make it weak or less complex to reduce tokens

[quartz coral]

no they can change the system prompt

to make it think "more efficently"

they dont have to reduce the weights

[topaz epoch]

I just need return of opus fine

I'm happy with that

[fast dew]

Why r people overrating opus so much

[quartz coral]

for most people it works out

[fast dew]

So does sonnet

[quartz coral]

each person has their own favourite model

[topaz epoch]

Opus is way more intelligent than others

[fast dew]

Use sonnet thinking

[solar venture]

I'm now looking for another website like Arena. Do you have any suggestions?

[quartz coral]

just tell them to try other models or maybe theyll find other alternatives to continue using theirs

[topaz epoch]

Try this

Car wash is 100m away from home. I need to wash my car should i walk or drive.

You get which is best

No competition of Opus 4.5 thinking and Opus 4.6 thinking

[modest topaz]

I guess they are playing with us, April fool prank or something and they will bring back Opus I hope. 🥲

[fast dew]

Literally the same results you'd get from opus bro

[quartz coral]

they probably will

[solar venture]

Really

[quartz coral]

no one knows yet

[fast dew]

Holy cope

Js wait dawg

[primal orbit]

I thought it was Anthropic and other AI companies providing API's for arena.ai in exchange for data in conversations.

So arena.ai wasn't actually spending money on chats.

[fast dew]

Bro went quiet

[plucky sparrow]

I think Anthropic realized people weren't subscribing and using Arena.AI instead 🥲

well that, and I think there were a bunch of abusers

[solar venture]

Like we.

[plucky sparrow]

yeah.. so it's probably not coming back 🥲

[fast dew]

It's one of the most top downloaded apps

???

[neat echo]

Yo

[primal orbit]

TBH direct chat option was very generous. It should have always been just battle mode.

[plucky sparrow]

probably not a lot of real subscribers though. a lot of people are free users

[plucky sparrow]

yeah.. we were lucky to have it while it lasted

[fast dew]

I'm sure there r way more subscribers than people that use arena

[plucky sparrow]

until people found out you can use arena.ai and get more opus use than paying $20/month 😄

[primal orbit]

We were also lucky to have opus 4.6 on day 1 here in direct mode.

[fast dew]

Most people don't even know about arena lol

[plucky sparrow]

personally I think the abusers were the main problem

[fast dew]

Wdym abusers

[plucky sparrow]

I suspect there were bots. but it's only my suspicion

[quartz coral]

u can circumvent captchas

yeah botting can be really a thing

[plucky sparrow]

yeah

[topaz epoch]

You used thinking 😭

[fast dew]

Lowkey they should just do paid subscriptions for direct chat

[topaz epoch]

Good work

[fast dew]

??

Yes

[primal orbit]

yeah, like open router.

[topaz epoch]

Btw thanks bro

@fast dew

[fast dew]

Your welcome

[fast dew]

Or they could decrease the rate limits

[topaz epoch]

Thinking versions are best

[topaz epoch]

Exactly

[fast dew]

Like 5 to 2 per hour or smth

[plucky sparrow]

that's actually not a bad idea, it'd be good anti-bot

[fast dew]

I js got a feeling they're gonna like nerf the opus model when it comes back

[topaz epoch]

I can't find sonnet 5 thinking ?

[plucky sparrow]

a lot of freeloaders would complain but better than not having it at all

[primal orbit]

it's way below. type sonnet in search

[topaz epoch]

Only 4.6 is showing

[quartz coral]

theres no version 5 that anthropic released

what are you on

4.6 is the latest one

[vapid mortar]

Opus 4.6 is gone???

[primal orbit]

I used Opus search to analyze the way Iran war is going. When the oil went to 100 it said it's a trap and it will get back high again. And it did.

[elder solar]

Is this the end of free ai access?

Since yupp ai is being gone soon too

[fast dew]

U literally have access to every other model except opus

[elder solar]

I was more focused to opus though

[fast dew]

Js use sonnet thnking bro

[elder solar]

I did find 4.1 opus be more aware than any other model

Not the best one i know

[topaz epoch]

But see bro can't upload picture and pdfs in thinking one

[elder solar]

But I mean as in personality aware

[topaz epoch]

In Opus that was allowed

[elder solar]

I had to use math for see which ais are really personality aware

Or personality that looks too human

[elder solar]

The only option for free 4.6 opus is Google antigravity

But that's for only pc

[quartz coral]

maybe because the system prompt of the ai thinking it was roleplaying as a character called "Claude" featured a ton of personality prewritten

in it

unlike the newer ones

[topaz epoch]

Ahh 🙂

[primal orbit]

I think if you're not doing a lot of coding or work, paying for opus API should be ok?

No way I use 1m output per month

1m is 25 usd

[elder solar]

qwen3 next 80b a3b instruct is another model that is personality aware too

[quartz coral]

hm

yeah then its some quirks inside of those

[elder solar]

I had to list which models feels the most human brain

[quartz coral]

idk really how to explain it lol

[elder solar]

And nvidia nemotron nano 30b seems to hit the same spot too

[undone saffron]

NvidiaTeslaT4

[elder solar]

The only two that are still on arena website

[elder solar]

That's a gpu

[elder solar]

The only ones that would be always personality aware are finetunes

[ashen plaza]

so wait is gemma worse than general gemini? ive been trying to use it for creative writing but ive been noticing quite a few issues compared to gemini. i assumed since it was newer it would be better. or is there something im missing?

[plucky sparrow]

hm, i thought of one way to rid all but the most complex bots a while back

but most people not willing to implement it

a constantly changing, unique captcha

it doesn't even have to be complicated. it just needs to be different and not recycled

[elder solar]

Gemini is just confusing for me

[undone saffron]

I know, I only said that because you said “nvidia” and I had that word copied

[elder solar]

Last time i tried get gemini talk about a youtube music video, and instead it just gave other unrelated songs

[plucky sparrow]

e.g. don't use cloudflare, don't use google recaptcha, but create your own, and change it randomly

[primal orbit]

gemma is a smaller model, therefore much less capable.

[elder solar]

It always go off topic weirdly

It's like they did feed gemini with alot useless pop culture

[plucky sparrow]

with the age of LLMs, it actually shouldn't be hard to create new captchas weekly

[quartz coral]

gemma4?

[ashen plaza]

well. that explains it then. Dang that sucks, i got excited as it has a lot more free uses on ai studio but if its smaller than man ;-; oh well back to gemini

[quartz coral]

yeah gemma is an ai meant to be ran locally

[ashen plaza]

yeah

I didnt know that

[quartz coral]

BUT atleast the new 4th one has some stuff taken from gemini 3

[plucky sparrow]

yeah gemma4 is meant to be able to run on phones

[quartz coral]

so its way better than gemma3

plus it has a unique thing against other open source ais is that it has native vision (image analysis) and audio analysis

so its a pretty decent model for general tasks

[elder solar]

Step 3.5 flash seems to be great for creative writing too

I've got the model noted inside the 'personality aware' list of models

Gemini 2.5 pro seems to be really decent for writing too

[elder solar]

Audio analysis seems pretty chaotic for me

It always gonna think such song does exist and published by such artist

When the song file is from unreleased takes

[echo dome]

hcaptcha and open source captcha exists

[elder solar]

It should just tell me what details i said i need on the prompt

[ashen plaza]

TBH 3.1 is my favorite for creative writing outside of the very limited amount of useage. I suppliment it with 3 flash. I dont really care for 2.5 feels bad to me now after using 3.1 and 3 for so long now

[elder solar]

It should just tell me what details i said i need on the prompt

It should just tell me what details i said i need on the prompt

Not useless random labels

[echo dome]

happy birthday (just checked your about me) you're same age as me

[elder solar]

Dang why did i repeat

[static heart]

Why are the best Ai Models removed?

[echo dome]

i keep accidentally double tapping reaction

[elder solar]

Oh tysm

[quartz coral]

yeah its in the workings i guess

[elder solar]

Any other model than gemini?

[ashen plaza]

for what I use? occasionally grok and occasionally claude but both feel ike they suck for what I want

[elder solar]

Just those?

[elder solar]

Just those?

There's alot proprietarys

[ashen plaza]

IDK what else I would use

[elder solar]

There is kimi too

[ashen plaza]

I haven't found anything that feels as good as gemini to use. on top of being. ya know. free

[elder solar]

Kimi is supposedly trained for more humanized text writing

[ashen plaza]

I stopped using kimi early on and I dont remember why

[elder solar]

Deepseek?

[ashen plaza]

i didnt know deepseek has a website

[elder solar]

You must be really outdated rn lol

[ashen plaza]

litterally everything ive seen about deepseek talks about the api only

[long cargo]

I just found out that coding with Sonnet is such a painful experience. Whenever I put a code generated by sonnet-4.6 or claude-sonnet-4-5-20250929 and try to compile it, I keep getting compile errors on my screen. Woof.

[astral jetty]

I found kimi to be really bad

[lapis imp]

Any recommendations models for creative story writing or roleplay? I want to test and see

(sorry for interruption)

[astral jetty]

I honestly think Gemini 3.1 is a decent amount better than opus for general daily and creative uses, but opus was the only option for making like long form content

[atomic lagoon]

These must be anthropic models

[atomic lagoon]

Creativity opus is certainly better but general use 3.1 is better for sure

Opus is just so expensive

Hence why its a one shot model

[long cargo]

well, are there any modles that we can use to get good angel script codes?
opus was the best, now i'm consernd about since we don't have access to opus models anymore.

[elder solar]

Step 3.5 flash?

[atomic lagoon]

Codex 5.3

Best after opus

Its heavy backend though thats all

[long cargo]

do we have access to Codex 5.3with a free account?

[atomic lagoon]

Yes

On arena

[long cargo]

o, i should give it a shot

[fossil socket]

Yeah I'm not using the website for the time being then

Disappointing update

[midnight wagon]

Do you see Opus 4.6 on this page: https://arena.ai/text/side-by-side It was there yesterday, but today it's gone.

[fossil socket]

I'm afraid not

[light sleet]

Rip Opus

[fossil socket]

How most users are feeling rn after the update:

[midnight wagon]

Is there any news on why Claude Opus 4.6 is not available on the page https://arena.ai/text/side-by-side?

[fossil socket]

#announcements

[tiny jolt]

Well... RIP

[fossil socket]

Istg do y'alls not check that

[echo aurora]

There were some models that were removed from Direct and Side by Side modes

[midnight wagon]

I saw disable-opus flag in the local storage

[merry reef]

Why does it always say error when I upload image ?

[echo aurora]

Would you mind asking in #ask-here ?

Oh you did

Should get a response from the bot soon.

[merry reef]

Yea but no reply there or am I to tag something?

[light sleet]

Could they come back or no more voting for newer models of opus and other?

[echo aurora]

Yeah that's odd. Can you ask it again, but this time don't tag the channel?

[echo aurora]

The announcement mentions:

we intend to bring these models back in a way that’s more sustainable when possible.

[merry reef]

Ok got a reply but it didn't really help

[light sleet]

Probably when Opus 5 releases 💀

[fossil socket]

Anyways since apparently the website now no longer includes the other opus models

Yeah I'm dropping the website indefinitely

Until they revert this asinine update

[ashen plaza]

WOW deepseek feels like it kind of sucks. Im running a test using a complicated writing scenario that Gemini completely nailed and it's struggling beyond belief. It's confusing things, mixing them up, forgetting the basic rules of the scenario within 3 messages. Meanwhile Gemini using a mixture of 3.1 and 3 went for 20-30 messages straight without making a single mistake and remembering the rules of the scenario.

I'll give deepseek a few more tries but it's quickly hitting a is this supposed to be good feeling for me. Maybe it's just. Running poorly right now or something? Idk

[echo aurora]

Responded in the thead, we'll chat there.

[merry reef]

God bless

[echo aurora]

I'm sorry to hear that. We hope to bring them back one day. It is your decision to make though, and I understand that.

[outer zinc]

claude opus models got removed from direct chat?

[proper quiver]

yeah

now its only the sonnet model

[fossil socket]

Yes the disappointing model cousins

[proper quiver]

yeah thankfully i can code with the sonnet models instead of the opus models

[midnight wagon]

RIP arena

[proper quiver]

(im working with roblox luau)

[atomic lagoon]

Anyone know which stealth model emu is?

Or clue

[jaunty mist]

the enslopification of lmarena

deleted my account at this point

[wicked talon]

Just code with qwen atp

Or kimi

Probably the best models left to code

Or grok

[atomic lagoon]

Its honestly crazy people are hating so much over this lmao. I couldn't even use opus 4.6 half of the time because the site would cause an error, there were high rate limits as well. Clearly they couldn't handle it, people are so weird for complaining about this

[atomic lagoon]

Qwen sucks kimi is alright

But like

How do yall forget that codex is there

Sure GPT 5.4 and gemini 3.1 isnt but codex is

[fossil socket]

For creative writing opus 4.6 was the best

[wicked talon]

Qwen isn't bad at all lmao

[atomic lagoon]

Oh 100% but still, a new model will come specialized in that believe me

[atomic lagoon]

I was testing it yesterday, its not very good

And the benchmarks they used for it

[wicked talon]

It's amazing for me

It's 3.6 plus is close to gpt 5.4 in my opinion

[atomic lagoon]

Were benchmarks against last generation models and it still didnt perform great against the LAST generation ones

[atomic lagoon]

Well yea because gpt 5.4 isn't the coder model lol

Test it up against codex 5.3

[wicked talon]

The last qwen coder was qwen 3 coder

[jaunty mist]

this
especially after google lobotomized 3 pro after the release of flash

[wicked talon]

Not a fair comparison

[atomic lagoon]

Right

[wicked talon]

Your comparing a specialized coding model to a general purpose model?

[thorny schooner]

Hope you all are doing good

[atomic lagoon]

Then why are you coding with a general purpose model? Especially when there are models that specialize in code

[wicked talon]

Wait for qwen 3.5 coder then try

[atomic lagoon]

Sure, I wouldn't be surprised if its good

But we will only see when it happens

Either way though

Especially with this claude code source leak

Stuff is going to get big

And better

[wicked talon]

The thing is using multiple ai isn't good for memory and cost

Like you could tell 1 ai one thing and the other ai doesn't know

[upbeat flint]

qwen 3.6 any good or about as bad as 3.5?
also still not seeing opus 4.5, wasn't it supposed to stay?

[wicked talon]

3.6 is a little upgrade

Not too noticeable

[atomic lagoon]

Right but if you're coding you would use a coding model because majority of them nowadays have websearch as well

[fossil socket]

Honestly hoping there'd be another worthwhile creative writing model but for now I'll just wait for Opus then

[wicked talon]

Qwen 3.6 plus holds up well against gpt 5.3 codex

[atomic lagoon]

Models get base trained off of fineweb so it understands everything off the web to begin with then comes the absurd amount of datasets of code to give it

[atomic lagoon]

For frontend I'm sure

Nowhere near as good backend

Codex 5.3 is a backend model

[upbeat flint]

I mostly ran into infinite thinking on qwen and it wasn't very impressive for translation work

[wicked talon]

Considering it's like 6x cheaper and a general purpose it's really well

[atomic lagoon]

This is cost not actual production

[wicked talon]

Score + cost

[atomic lagoon]

Run the two in battle mode off the same prompt, I will in a minute

[wicked talon]

5.3 codex ain't on battle mode

[atomic lagoon]

Oh damn

Or side by side oops

[wicked talon]

I'm gonna side by side qwen 3.6 max preview Vs gpt 5.3 latest

Making random htmls etc.

Like games

[atomic lagoon]

Hang on ONE second

My mind has been changed

Not between codex and qwen

I mean just qwen

https://019d55bc-b5d6-73ac-b740-c56af6a27afd.arena.site/

[silk stirrup]

I said it before somewhere, but instead of removing flagships, it might be better idea to remove useless models nobody uses

[atomic lagoon]

I wanted some anime type site

Using jikan

API

[wicked talon]

Quite good

[atomic lagoon]

It made something I've never seen

[jaunty mist]

that being what? opus 5 lol? to be honest, the only thing that came close to 4.6 was gemini 3 pro on its early release (or so it felt)
then, again, this entire website became useless to me as well, since the model that i use for general stuff is gemini and ai studio is just lightyears ahead from arena in every way possible

[wicked talon]

Bruh I support qwen for a reason

[upbeat flint]

how so, just temperature etc settings for gemini or?

[wicked talon]

Gpt is lazy asf

[frosty lava]

Qwen 3.6 for the compute power they have they've done a really great job

[atomic lagoon]

You have to prompt it right with gpt

But thats anything

[wicked talon]

Qwen works when your vague lol

[atomic lagoon]

AI is trained to pick the easiest route

Actuallt

Not purposely trained to take the easy route

They literally just do that automatically

[frosty lava]

small companies might actually beat the top tier huge one

[wicked talon]

Lemme send the script it produced

[frosty lava]

the gap is closer and closer

[atomic lagoon]

Indeed

[undone saffron]

I'm waiting for an uncensored version to come out, plus it uses fewer tokens and so on

[frosty lava]

and if you look at the price the difference is crazy

[wicked talon]

Here’s a simple Cookie Clicker-style HTML you can save as index.html and open in a browser:

<!DOCTYPE html>
<html>
<head>
  <title>Cookie Clicker</title>
  <style>
    body {
      font-family: Arial, sans-serif;
      text-align: center;
      background: #f5e6c8;
    }
    h1 {
      margin-top: 20px;
    }
    #cookie {
      width: 200px;
      cursor: pointer;
      transition: transform 0.1s;
    }
    #cookie:active {
      transform: scale(0.95);
    }
    button {
      margin: 10px;
      padding: 10px 15px;
      font-size: 16px;
    }
  </style>
</head>
<body>

  <h1>🍪 Cookie Clicker</h1>
  <h2 id="count">Cookies: 0</h2>

  <img id="cookie" src="https://upload.wikimedia.org/wikipedia/commons/7/70/Chocolate_Chip_Cookies_-_kimberlykv.jpg">

  <div>
    <button onclick="buyUpgrade()">Buy Auto Clicker (10 cookies)</button>
  </div>

  <script>
    let cookies = 0;
    let autoClickers = 0;

    const countDisplay = document.getElementById("count");
    const cookie = document.getElementById("cookie");

    cookie.onclick = () => {
      cookies++;
      updateDisplay();
    };

    function buyUpgrade() {
      if (cookies >= 10) {
        cookies -= 10;
        autoClickers++;
        updateDisplay();
      }
    }

    function updateDisplay() {
      countDisplay.textContent = "Cookies: " + cookies;
    }

    setInterval(() => {
      cookies += autoClickers;
      updateDisplay();
    }, 1000);
  </script>

</body>
</html>

[frosty lava]

using qwen cost nothing compared to using opus or gpt

[atomic lagoon]

Right but I'd trust GPT to do a job for a software development over qwen

[wicked talon]

qwen

[frosty lava]

in front end its really good

yh

gpt lacks frontend

[wicked talon]

havent tested it though qwen might be bugged

[frosty lava]

opus cost too much

[jaunty mist]

(generally) higher ratelimits
editable system prompt
user/ai messages are fully editable
prompts can be copied or even branched
safety filter can be disabled

[frosty lava]

and honestly right now i'd rather use qwen 3.6 than opus

[undone saffron]

Interesting name

[frosty lava]

its a big companies and they've bought alot of compute power to achieve this model

[wicked talon]

microsoft copilot is quite good at frontend

thats claude i think though#

[frosty lava]

is it a model or does it use another model

i think it just use claude or something

[wicked talon]

uses multi models

gpt and claude i think

[atomic lagoon]

At the end of the day, its always just best to prompt better because really this isnt great, but with the half decent prompt I did for that anime streaming site it made it beautifully

[frosty lava]

yeah

[wicked talon]

gotta remove the top i got that issue

[frosty lava]

deepseek v4 ive been waiting it for a very long time

[wicked talon]

me too lol i cant wait

[frosty lava]

when it come out it better be a huge improvement

[atomic lagoon]

It will be trust they've been releasing new and improved concepts, its gonna be a breakthrough

[atomic lagoon]

I see

[frosty lava]

they might beat gpt and claude with this one

[atomic lagoon]

Qwen also thinks longer than codex lol

[frosty lava]

if its a breakthrough

[wicked talon]

for a 8 word prompt it aint too bad

[atomic lagoon]

Its very hard to beat claude but realistically they could

[atomic lagoon]

Nah it isnt bad for that

[frosty lava]

i also love how qwen is fast

[neat apex]

like, they already did with Qwen 3 Coder that time

[atomic lagoon]

Heres what codex just made me, I am waiting for qwen to finish

https://019d55db-3ac6-7920-9fa3-5730ffa27141.arena.site/

[frosty lava]

it doesn't take 10 second to start thinking

[neat apex]

it felt like anthopric almost went bankrupt due it lmao

[wicked talon]

why does nextdns flag arena bruh

[neat apex]

while, yet now, most of tokens arent spent on code at all

[frosty lava]

honestly claude is good but the price make it useless almost

[wicked talon]

ohh its cause of new domain thing

[atomic lagoon]

Slower than codex haha, Codex got done in like 1-2 minutes for my prompt and codex is still going 8 minutes in

[frosty lava]

you can have a good subscription with claude and in few days your at max usage

[frosty lava]

yes but its because it thinks more than codex

if you look at how fast the thought process

[wicked talon]

if i pay for a model i want unlim usage

[atomic lagoon]

Its a one shot demon thats why, highly expensive unfortunately

[frosty lava]

you'll see

[neat apex]

unlike Codex, Qwen is not the best planner, actually i would say the worst by seeing it at all

[atomic lagoon]

But thats the thing, codex has better quality overall and thinks less I'd call that a win

[neat apex]

even worse than minimax, but idk about 3.6 yet

[frosty lava]

that's impossible but its almost infinite with any companies that's not anthropic or gpt or gemini

with small companies you can have ALOT of usage

for much slower price

it seems infinite

[wicked talon]

qwen is almost unlimited for chat

[frosty lava]

yeah

[neat apex]

Codex does things more dry than Qwen too, ya?

if 5.4 perchance added some significant charm is a detail

i guess its yet dry, besise more well done now

[atomic lagoon]

Dude exactly or like 200+ requests in a day because I aint gonna use all of that but also I have projects I want to work on and claude makes that impossible not having high requests

[wicked talon]

only ever got rate limited on deepseek for 2 mins

[frosty lava]

i like codex but i have to admit in frontend development its not good yet

[atomic lagoon]

Ehh it depends on how good your prompt is really

[wicked talon]

true

[atomic lagoon]

Thats what I've said for now, its a backend demon for sure. But then it just made me this and its pretty cool https://019d55db-3ac6-7920-9fa3-5730ffa27141.arena.site/

[neat apex]

i can get it

if you ask a lot, Codex manages to do it, while Qwen suffers and be more dry

[frosty lava]

i think they said on X that they gonna fix it (make it better at frontend)

if they actually do it

it will be a huge win

[atomic lagoon]

Now qwen is finally done, @wicked talon which would you prefer the claude one or the one qwen just gave https://019d55db-3ac6-77ad-9ecd-086f26385de5.arena.site/

Both same prompt

[proper quiver]

should i tell it to make a website in React?

[frosty lava]

for this one especially the gpt one is the best

[proper quiver]

should i tell qwen to make a site in React?

[atomic lagoon]

I think it just needs better prompting for frontend, more in depth explanation for the frontend design

[wicked talon]

cant see any being blocked for malicious stuff

[frosty lava]

but from what i tried qwen 3.6 is better at front end overall

[atomic lagoon]

Bruhhh

[frosty lava]

yeah i guess

[wicked talon]

lemme go on my phone and see

[proper quiver]

so i think in this context, qwen 3.6 is good at React

[atomic lagoon]

Qwen

Codex

[frosty lava]

yeah i think too

[proper quiver]

should i make it use radix primitives

[neat apex]

both are great xd

[wicked talon]

Qwen actually looks like a movie website

[proper quiver]

( https://radix-ui.com/primitives )

[neat apex]

but i guess codex did it more, idk

[wicked talon]

Codex looks like a advertisment website

[atomic lagoon]

Possibly but it depends on what you do I guess

[wicked talon]

Like compare it to netflix and qwen would be more accurate

[frosty lava]

try making a small 3d game

[neat apex]

i cant proof Codex or Qwen did it better than other

[frosty lava]

for example

[atomic lagoon]

The prompt was meant to have that design though, lemme send the prompt

[wicked talon]

Kk

[neat apex]

fine, lets see

[atomic lagoon]

Got this from gemini thats how I get my prompts lmao

Task: Architect and code a high-performance, responsive Anime Discovery Platform using the Jikan API (v4).

Design Requirements (The "Anti-Basic" Mandate):

Visual Style: Avoid Glassmorphism entirely. Instead, implement a Gothic Base or High-Contrast Gothic Grid aesthetic. Use Gothic Style typography, sharp borders, and a gothic dark, non-traditional color palette (e.g., Deep Charcoal with Acid Green or Cyber-Purple accents).

Animations: Use Framer Motion or GSAP for staggered list animations, hover-triggered layout shifts, and smooth page transitions.

Unique UI Element: Include a "Quick-View" sidebar or a Draggable Carousel for "Currently Airing" shows that doesn't rely on standard Bootstrap/Tailwind primitives.

Technical Requirements:

Stack: React with Tailwind CSS or Next.js.

State Management: Use TanStack Query (React Query) for the Jikan API calls to handle caching, loading states, and pagination without "easy-way-out" useEffect hacks.

Feature Set: Implement a robust search bar with "Search-as-you-type" functionality, a "Top Anime" landing grid, and detailed modal/page views for individual series including genres, scores, and trailer embeds.

Constraint: Write clean, modular components. Do not simplify the logic; ensure the code handles API rate-limiting or empty states gracefully.

[neat apex]

but its a clear scenario Codex did it less dry than Qwen

[wicked talon]

Omg qwen is still at the fuhing code I sent like 1 min ago

Really designing it well

[atomic lagoon]

Qwen was coding for about 12 minutes in the prompt I sent

[wicked talon]

Bout to be 2k lines type shi

[frosty lava]

but in my opinion the purpose is to be able to create what you want without having to do such prompt

[wicked talon]

It overcodes I think

[frosty lava]

at some point ai will be able to do it

LLM*

[wicked talon]

Ok uhm I think gpt 5.4 is dumb

Tell me why it produced a prompt

When I asked for html

[atomic lagoon]

Could be the case because unlike claude it doesnt think too much, it just codes too much or is slower at the coding end

[atomic lagoon]

LOL

[wicked talon]

I misspelled with but surely it could recognise it

[sand fulcrum]

it snot

[atomic lagoon]

wih

[wicked talon]

besides the banner and 2 useless buttons in the shop qwen did quite a good job

[sand fulcrum]

what is this for

oh qwen

[wicked talon]

just a html cookie clicker#

you can remove the banner if you delete the top line before code lol

but yeah this vs gpt 5.4 qwen obviously won

i mean gpt 5.4 produced a script

[atomic lagoon]

GPT has never really been the greatest at coding though, only codex really so it makes sense

[wicked talon]

wish codex was on arena to compare

[proper quiver]

i put this into claude

[atomic lagoon]

Yea me too

[atomic lagoon]

Hell yea, lmk how it turns out

[proper quiver]

it broek

[atomic lagoon]

Oh boy

[atomic lagoon]

Yup not surprised

Arena just has a problem with claude in general

Always breaking

[proper quiver]

it worked fine if i gave it some roblox code

[echo aurora]

Will look up this Trace ID and let you know.

[wicked talon]

oh lmao

i wasnt using code arena

[ashen grove]

Is claude opus better than qwen?

[wicked talon]

debatable

[atomic lagoon]

Opus is 100% better than qwen

But no way to tes tthat now

[wicked talon]

not for general

[ashen grove]

wasn't opus like lagging and bugging out whenever it thought for too long?

[ashen grove]

i mean for coding obviously

claude is like top for coding

[proper quiver]

and i cant use opus just for coding my roact

[ashen grove]

but like qwen is like build for coding too

[inland hornet]

The opus is missing , what happened??

[atomic lagoon]

See THIS is what I'm talking about qwen

[ashen grove]

remoevd

[atomic lagoon]

Ah yea

This is codex and I think the qwen design is much more unique

[wicked talon]

im comparing these bad boys with that script u sent

[atomic lagoon]

The codex app is just missing something

[atomic lagoon]

Haha

[ashen grove]

Qwen or claude sonnet

pls someone

[wicked talon]

qwen

[atomic lagoon]

Oh damn its an instruct model

[wicked talon]

lol why

[atomic lagoon]

Its called an instruct model

Which means when training a smaller model this instruct model feeds the newer model being trained its info

[wicked talon]

lmao

[inland hornet]

Which one is best for pattern recognition and prediction

[wicked talon]

https://019d55ef-e34d-701f-a27e-0f51460ca64e.arena.site/ qwen

[ashen grove]

uhh how much does arena spend on us using all this paid ai

[wicked talon]

https://019d55ef-e34d-777f-964c-b393f0a38f5e.arena.site/ gpt

who is better

[ashen grove]

@wicked talon how much does arena spend on us using all this paid ai

[rigid pasture]

I found a way to bypass the limit on infinite generation chats and want to see if it works for everyone. Is it okay to post GIFs in here? I'm gonna record a video to show you guys how

[wicked talon]

Alot but arena is funded by investors

[atomic lagoon]

I dont like either, but codex is closer to the prompt just it looks like a damn history paper

[ashen grove]

How does the investors even make money

how does arena make money?

low key i dont see no paid option

(which is great btw)

[echo aurora]

This is being caused by rate limit. Arena can be rate limited by the model provider. Can learn more in this article: https://help.arena.ai/articles/8931786544-arena-how-to-rate-limit

[wicked talon]

I dont think they do lol

[ashen grove]

And will arena soon support api keys

[ashen grove]

oh

[wicked talon]

they dont

[ashen grove]

ohh

[wicked talon]

they will be sued

[ashen grove]

wait what why

[wicked talon]

unless they get licensed by every ai

[ashen grove]

oh

[wicked talon]

they use

probably

[echo aurora]

We have an evaluation product, can learn more about it here: https://arena.ai/blog/ai-evaluations/

[ashen grove]

is pineapple a app

[thorny schooner]

It is value investment where you invest in something for other value that's not pure money basically in this case basically help data be scattered for AI which in turn proves AI which might help their Investments if they have those in AI companies and if not i wouldn't be surprised if some of it is from Bill was foundation for those that are interested the AI itself

[wicked talon]

No real person.

[ashen grove]

he looks appy

[thorny schooner]

no lmao but feels like it at times

[ashen grove]

i swear he uses chatgpt for responses

[wicked talon]

he is just following scripts tbf.

on a arena.ai help page

thats what most support agents do at any company.

[atomic lagoon]

Thats been talked about a lot

[atomic lagoon]

LMAO

[wicked talon]

Arena is not gonna do it as it will lead to legal complexity

[thorny schooner]

A bit too much sometimes for my liking but fair enough ( to be clear when I say that I get why you would make the help paige it's just you know sometimes it can feel with like if it happens way too often i or it is always word to words you get what I mean)

[whole aspen]

I thought Arena was free to get models from manufacturers who wanted to test model rankings...

[thorny schooner]

That's like asking for a needle in a haystack lol

[wicked talon]

No.

Its all paid for i believe

[ashen grove]

but if using a magnet

[atomic lagoon]

They will 100% do it at some point, think of pollinations ai, its possible and really is likely

[ashen grove]

also does not arena support html files or txt files?

[undone saffron]

@ashen grove

is there any other website other than arena (which previously) that lets you use claud eopus 4.6 without limits
Yes, google is your friend

[ashen grove]

wat

[atomic lagoon]

Yupp.ai has opus 4.6 just not in direct chats

[ashen grove]

gonna have to go back to claude

[ashen grove]

yupp cuts off claude when its thinking for too long/writing /coding for too long.

[wicked talon]

Are they not gpt 4 etc?

older older models

[atomic lagoon]

Arena did the same thing 😂 idk any others though

[atomic lagoon]

Sure but no legal troubles

[wicked talon]

GPT 5.3 etc would be more complex

[rigid pasture]

Welp... I just lost access to Opus. It was good while it lasted

[burnt sinew]

vc funding

[atomic lagoon]

Nah, all it is, is AI lol

API mb

[burnt sinew]

like most other ai companies

[sand fulcrum]

yupp.ai?

[sand bluff]

the interesting part is opus 4.6 is still avaliable in battle mode if you get lucky

[whole sundial]

but yupp just shut down, why are you still talking about it like it still exists

[sand crown]

Does anyone have issues with token usage limits? I send four one sentence prompts to Grok 4.20 multi agent beta 0309 and it hits the limit

[rigid pasture]

Reset the page, say hi to GPT 5.4, and then try using your AI again — I know, it's very specific.

[sand crown]

Thank you

[undone saffron]

@quick jackal@echo aurora

[ashen grove]

somebody ban @cloud fox he is a gambling advertisement bot

[undone saffron]

Done

[rigid pasture]

@echo aurora

[spark linden]

People who used claude for text and story telling, what are you guys going to use now? I'd like recommendations

[ashen grove]

??? text and storytelling??? i though it was 99.99% coding

the remaning 0.01% for providing context

[rigid pasture]

Men

The guy created two accounts just for that

[ashen plaza]

TBH or they fell for a scam at the same time. both accounts had joined in november last year. they could have both fallen for something a bit ago. but idk

[small vigil]

are there seriously no alternatives out there for claude opus anymore

[sand fulcrum]

just use max

or wait for the models to come back

its not getting fully removed eventually

[rigid pasture]

Yeah

[small vigil]

theyll come back?

[rigid pasture]

probably

[sand fulcrum]

yeah

[atomic lagoon]

<@&1349916362595635286>

[sand fulcrum]

we intend to bring these models back in a way that’s more sustainable when possible.

[small vigil]

oh nice

[sand fulcrum]

thats what the announcement said

[atomic lagoon]

Yes

[small vigil]

shoot i just joined i noticed it was gone on the website

[proper quiver]

reported as scam

[zenith steppe]

poor kid, he got his acc hacked

[proper quiver]

@ivory anvil got reported

[rigid pasture]

Men

5 accounts with the name Maycon Hector

[proper quiver]

the scamming or defrauding button seems very shiny today

[undone saffron]

I wonder what token grabber program they might have run

[sand fulcrum]

no wonder my main acc got

logged out

because of that token

[wicked talon]

Qwen for frontend gpt (codex) for backend

Trust

[undone saffron]

POV: A kid searching in google for first time

[plucky sparrow]

What's this?

[wicked talon]

Wait what lmao

[proper quiver]

i told gpt 5.4 to make it use material

https://019d5624-210c-7622-bbf9-2890e198f866.arena.site/

[rigid pasture]

The website is really nice

[mighty stag]

Hello, I just noticed that Claude opus models were all gone, why this happened? I know this would temporal, though

[atomic lagoon]

True that

Airing currently

I cant wait for Mushoku tensei

[echo aurora]

Some models have been removed from Direct and Side by Side. More info can be found in a recent #announcements

[mild spade]

will it be returned☹️☹️

[echo aurora]

We want to when we can provide it in a sustainable way

[subtle rose]

What do you mean?

[gloomy onyx]

woah, here's another C# fella

[subtle rose]

https://tenor.com/view/ronaldo-chug-chocolate-drink-gif-21574013

[blazing mulch]

hlo

[unborn robin]

is any feature on arena free and unlimited

[atomic lagoon]

They're all free and not really unlimited due to rate limits but thats it

They aren't going to bring it back, its genuinely just too expensive

[unborn robin]

but rate limits reset and are quiete large right

[atomic lagoon]

Depends on the model you use

[unborn robin]

gpt 5.4

[atomic lagoon]

Shouldn't be that bad

It depends on how expensive the model is which will determine the rate limit

[unborn robin]

what’s best ai rn for arena. also what’s different between using arena 5.4 and paid 5.4

[atomic lagoon]

So basically nothing really has bad rate limits

[atomic lagoon]

Wdym paid 5.4?

[unborn robin]

as in like paying for subscription

[atomic lagoon]

Ohhh

No difference, arena is just API

So arena is free

[olive spruce]

Yeah

[atomic lagoon]

Uses OpenAIs API in the site for it to work

Oh here he is

[olive spruce]

Lmao

[atomic lagoon]

The big Claude

[olive spruce]

Yessir

[atomic lagoon]

Left arena to chat in their discord server

[olive spruce]

https://tenor.com/view/flight-reacts-laughing-rage-bait-gif-8366393814504091925

I got tired of thinking so I left arena

[unborn robin]

y do they make it cost so much money when arena has free

[olive spruce]

Bruh
Arena is just a leaderboard and comparison website

[atomic lagoon]

Because arena pays openAI for the API requests using their model

[olive spruce]

Technically it’s not supposed to be used for personal projects
Technically

[unborn robin]

so how does arena get the money tho

[atomic lagoon]

It can be used for anything

[olive spruce]

Investors

[atomic lagoon]

Investors

[olive spruce]

Holy moly

[unborn robin]

ah so the users get the most benefit

just using paid ai for free

[olive spruce]

Yeah

Yes

And they get our data

[unborn robin]

r there any other websites that have paid ai for free

[olive spruce]

For free

[toxic verge]

https://capturingcaptcha.com/

[olive spruce]

None as good as arena lol

[patent bane]

AI is expensive, nothing lasts forever, remember that

[atomic lagoon]

Yupp.ai

[unborn robin]

it’s down

[atomic lagoon]

Yo thats crazy

[unborn robin]

do u guys all use arena

[atomic lagoon]

Yupp was here yesterday wild that its gone now

[unborn robin]

wait just to confirm arena pays per token from openai etc right

so it’s lowkey expensive

[atomic lagoon]

Nobody needs to worry about their expenses as long as its working

And gpt isnt that expensive

[olive spruce]

And almost all ai companies spend more than what they have lol

Lmao

[gloomy onyx]

kind of

they sure have some kind of discount because they rely on API paid plans which are tailored to fit big enterprises such as Arena.AI

but it's still extremely expensive

[olive spruce]

I thought your name was advertising

[undone saffron]

Me:

[olive spruce]

💀

[gloomy onyx]

yeah a lot of people say that

[olive spruce]

Lmao

Well gg’s to Claude for now
Let’s just use sonnet 😭🙏

[undone saffron]

I just accidentally discovered a vulnerability that lets you upload any file and have the AI read it

500k of text in 1 message

[olive spruce]

💀

[undone saffron]

But everyone can use you
Because you're Opus 4.6

[olive spruce]

https://tenor.com/view/son-im-crine-crying-emoji-anthony-mackie-gif-3482393872062328307

[olive spruce]

If you don’t mind
I would love to know about this in dms
Hehe

[frozen zealot]

actually nobody can use opus 4.6

check site

[olive spruce]

https://tenor.com/view/flight-reacts-laughing-rage-bait-gif-8366393814504091925

[olive spruce]

Yes that’s why I came here

[undone saffron]

Literally a meme

[undone saffron]

Sadly I can't
But if they ever implement [this](#1469637546828103741 message), you'll be able to do it just like I do

[olive spruce]

Lmao bet

Let’s hope for it

[desert pendant]

bro

😭

[undone saffron]

It would be funny if they put a character limit on the file you upload

<@&1349916362595635286>

[olive spruce]

Get your aaaah outta here

[olive spruce]

Is that a hint lmao 😭😭😭

[toxic verge]

It will be a thing

People have no limits and will push it to the most extreme

[undone saffron]

Better yet
Free pentesting

[toxic verge]

Ai safety always takes a back seat ;P

[proper quiver]

https://cdn.discordapp.com/attachments/1481306626865762418/1489844053783609375/attachment.gif

[heavy blaze]

@echo aurora nah we got @painapple

get it

bahahahhahahhahaha

[sand crown]

I can only send one sentence message and then the token limit is reached. This is so annoying

[short sluice]

ok so opus is gone on both my accounts

Though 3.1 and 5.4 are still there

[bleak lake]

What's opus 5.4

😭

[rigid pasture]

Gemini 3.1 and gpt 5.4

[bleak lake]

3.1 is also gone

[boreal anchor]

And we know when will return?

[rigid pasture]

They said that probably

[bleak lake]

till it gets profitable for arena

[bleak lake]

Those models are not profitable at all

[boreal anchor]

Dang it... I loved Opus 4.6 ;-;

[bleak lake]

yupp ai closing down as well

so yeah

[short sluice]

I'd say 3.1 is better for explaining things and acting like some people but 4.6 is better at making messages long and creative writing

[boreal anchor]

Exactly, Opus 4.6 is better for creative things...

[short sluice]

3.1 is much better at informality though

[boreal anchor]

I'm new to all the AI things, but, there isn't a way to fusion models?

[velvet furnace]

I join this discord because the claude 4.6 can't use now

[undone saffron]

#announcements message
#1489407244612145272 message

[velvet furnace]

😒

[unborn robin]

guys

is it possible to rename chats?

the only option i see is archive, also when u archive where can u find these chats

[storm granite]

nop

[gray isle]

you can't see them sadly

[gray isle]

it's my only LaTex Helper and now it's gone 😭

[patent bane]

not surprised

[gray isle]

CoDex kinda bugs me as it's not following my request as usual.

[odd geyser]

Why was the model opus deleted?

[humble crow]

I was surprised by that too; I searched for it this morning but couldn't find it.

[odd geyser]

He disappeared from me about a week ago. Maybe more.

I just don't understand why it was removed.

I know that it was removed as an experiment or something, but even after reading the article, I didn't understand.

[humble crow]

it wasn't removed until around 7:00 pacific time today

[dense sphinx]

Good morning people.

How are you holding without opus?

😕

[echo aurora]

I was done to maintain reliability. We'd like to bring it back when it's more sustainable

[humble crow]

is it an API-related issue?

[eternal pewter]

opus-4.6 was canceled—any alternatives?

[dense sphinx]

Hey pineapple.

[odd geyser]

But I could still use it for a while... I don't know what to call it.
I just went incognito and could choose the Opus of the model and chat with them. It was uncomfortable, but still a little gratifying.

[dense sphinx]

@odd geyser you used incognito?