#š¬āgeneral
1 messages Ā· Page 507 of 1
I feel like there were very few drops today
still waiting for pplx recap...
Will it happen the recap
Very little activity in the channel this very few drops
few people talking
no labs does not use sonar
hallo
what's your message stats?
way too slow in here this afternoon. I feel chatted out after the last couple of days. haha.
True
What can you say about Xiaomi MiMo V2 Flash?
It's awake
Have you used it?
oh no....I thought you were celebrating the arrival...not cursing the departure!
Not yet, just read announce
š¤£
funny either way. lol
is that the glm model?
Good evening, yesterday I accidentally tried to activate Perplexity Premium for businesses, and now it sends this red warning at the top when I open Comet (extremely annoying) can anyone tell me how to remove it thanks a lot
Delete the organization within your Perplexity settings
@wise nova you can stress Test it:
The numbers Mason, what do they mean?!! 10 23 68 111 8 7 7 47 53 23 63 92 15
Below are key recommendations from international bodies regarding paid parking, road tolls, congestion pricing, and the introduction of any charges for motor transport. The documents are grouped by organization.
Unlimited using?
How does an OpenRouter compare to Pplx?
I think I cooked it š
OpenRouter is not really a monthly subscription - you load credits and then use them at your leisure. You have to pay the actual provider API cost (plus a slight overhead for OpenRouter's take, if your queries exceed a certain amount)... so you miss out on some potential value based on what the providers offer directly, or what PPLX is able to negotiate. The responses are also contingent on pretty much only what the providers support, they don't use Perplexity's enhanced web searches, and they don't have anything like Spaces [as far as I know]... stuff like that.
Open router will be verrrry expensive.
Lol.
You won't be able to control token usage.
For equivalent usage, if you use a lot of stuff, yeah... if you try out just a few things - and are very mindful of your limits (you can set budgets in openrouter) then maybe it's more manageable... depends on what you're doing.
You might end up burning money without getting a satisfactory response.
Atleast that's how my experience was on poe.
What if I use OR only for the free Xiaomi model?
Well the model is free right now
Yeah, lots of people just use it for querying some free models.
But I still don't have the answer to The numbers Mason, what do they mean?!! 10 23 68 111 8 7 7 47 53 23 63 92 15
Gemini runs it better
Gemini flash is actually pretty good for free tier.
I agree
I don't think there's even any limit on it.
On AI Studio? Yeah, there is.
I'm talking about the app.
Oh, yeah - it's 'general usage' there - subject to throttling on demand, but most people just say it's practically unlimited.
I need one of those 8000 mah battery chinese phones.
But you can only use the non-thinking one with unlimited usage ('general access' subject to throttling)... the thinking version of flash is still not unlimited, which is the really good one. Still, the non-thinking one still scores slightly better than 5.2 non-thinking, according to benchmarks at least.
Honor Win with 10000mAh
Yeah. They added cooling fan too.
Is "thinking" the same as deep research or does it mean something else?
Where?
No it's usually not the same.
Just in general. Several seem to have the same terminology. Is it just deeper analysis?
@green crystal copilot deep research might be better than chatgpt deep research.
Thinking = reasoning. Deep research is more like a combination of reasoning and mass web scraping.
It varies a lot. They usually gather more data, and provide longer responses.. more recent deep research offerings even started embedding multimedia elements in responses. The idea is it's more like a research report. Though it's still well short of than I would usually call a report.
Though chatgpt is an exception where thinking outperforms deep research imo.
I think they probably just haven't updated Deep Research since 5.2 came out, but not sure.
ChatGPT used to have the best one, it's probably still up there at least.
So use case comparison- when would non-thinking be preferable to thinking? I can do non-thinking all by myself. Haha
It's pretty nerfed for the free tier now atleast. Lastest deep research were based on O3 and o4 mini.
I would say never, unless you are in a hurry.
Or you want to save tokens.
If you are api user.
I think the chain of thought reasoning still messes up chatbots sometimes, so if you're really just chatting or conversing / want quick generations. Nowadays, it's good for a lot of tasks... you'll probably get a feel for what you know it can do before having to switch over.
Which answer?
What if the bauble 1st place was awarded 1 month max and the rest 3 months, would that set a competitive scenario?
Good evening, yesterday I accidentally tried to activate Perplexity Premium for businesses, and now it sends this red warning at the top when I open Comet (extremely annoying) can anyone tell me how to remove it thanks a lot
That answer
Hi everyone! Iām a Perplexity Pro user and the founder of ECOHEAL (grounding and wellness products brand).
3 weeks ago(12.04) we submitted an application to join the Perplexity Merchant / Instant Buy program so that users can buy our products directly in chat, but we have not received any update yet( SILVER LINE, brand ECOHEAL)
Could someone from the Perplexity team please help check the status of our merchant application or point me to the right contact person for merchant onboarding?
Thank you!
hi
you can try actually clicking on it
and then cancelling etc
Whatās goin on? Perplexity has the talking system? Or text only?
talking too
Canāt really do it, it thinks too hard, aināt a convo
Is it down? Perplexity
No itās not down
i see
No
i haven't used it much
I'm getting errors "Sorry, something went wrong"
Try clearing your cache/cookies
just try again? or it's consistent?
I've already create a lot of other threads
which model are you using
Will try that
What platform are you on?
Tried various
lemme test a bit
Best to use the Perplexity app while on mobile
gosh still resets back to Best
Not for me
why use the web browser when on mobile?
security
Nope rip
That makes no sense
Baaauuble
Use the mobile app
Not how it works
Can't use connectors there
yes it is
if u have no software at all, u can't get attacked
balance is usability vs convenience of cuorse
Then why bother getting a phone?
Error is likely resulting from using the web browser. Please make a feature request for connectors on mobile #1358914210339360798
No issue on iOS either
Do you use the Perplexity App on your phone?
Do you use the Perplexity App on your phone?
I don't think you understood my question properly
Do you use the Perplexity App on your phone?
I'm asking you
not the person asking for help
why
Android
try log out and in
Are you switching apps while the response is generating?
They would have already done that after clearing cookies
I've used this plenty, it's just right now ig
Nah I get the error almost immediately
odd
Weird, perhaps make a #1140622008086970420
all models?
It shouldn't be happening at all, trying "various" models is enough
yes but more info might be helpful to troubleshoot
if it's only 2 or 3 models, that might point to those specific models being the issue only etc...never know with bugs
event commands have expired buttons now btw?
did anyone work out if Labs resets the counter after 30 days ?
or you need to wait for it to regenerate at ~1-2/day?
It's a rolling window, just like the other models
No, you can learn more here #š¬āgeneral message
how much?
3 euros bro
very strong performance as well
wow
coming close to opus 4.5
here we go
I tried
still waiting for 3rd party benchmarks though
still an issue?
Ive tried it
Yeah
sounds like u are frozen out...have u tried on windows?
same
Intergrated with claude code
their coding plan is a heck of a deal
Gave it custom mcp servers
And a lot of subagents
I would said with correct prompting it kinda feels outperforming opus
Ngl
I did not know Flux and Dalle are unlimited.. did someone tell that person who was asking about unlimited image gens a day or two ago?
The one who was generating cars and then saying they didn' look real?
This information comes from an old Perplexity FAQ. I am not sure about now
I haven't hit any image generation limits recently so I couldn't test it out
Will update my faq though Updated
i think we still not clear on whether a reset occurs imo
MiMo shows me very good results, maybe even better than 5.2 Thinking
what's MiMo?
Xiaomi MiMo-V2-Flash
link?
Xiaomi new model
https://openrouter.ai/chat?models=xiaomi/mimo-v2-flash:free
Modular Interactive Machine Intelligence
?
os
Aint no way this is real
It performs pretty good on AA benchmarks
at least for a company that makes every product imaginable
Isnt thag the old model
For a free model, that's pretty good
It just came out sooo
mhm
I'm still waiting for MiniMax M2.1 benchmarks
So many models
yep
I might cancel my claude max plan ngl
And just switch to glm
Its lowkey the same shit
And better limits
Honestly, either switch to GLM/MiniMax/Kimi or use Codex
Iam a big codex hater
MiMo lost its mind after two questions and spouted random characters... š¢
Lol, what has Codex done
When did you try it?
Hm
Not good for my coding at least
Many people use it for code analysis
But agentic workflow on codex sucked
It's been pretty good for me, not the best but not the worst either
Opus 4.5 is amazing, just the pricing
I ended up switching from my Claude Max 100x to just the Pro
and I've been using other OS models and codex more
I don't even use Pro to code now, just general things
Gotcha
Opus 4.5 uses up like 80% of my usage in one query lol
yees
Its pretty good from what I heard
But with thr existence of mcp server its kinda useless
Ngl
I guess it's just easier to onboard new users
Its easier to take 200 euros from people that dont have an idea
Hey at least you can teach opus to do shit in your browser
Like draw u stuff inside of a drawing tool that is fully web based
We havent seen thst yet ig
Yep
Could be cool
Mimo is hallucinating and makes mistakes
What better to set?
What is a presence penalty?
After the update, the text-to-speech feature disappeared.
Have you tried it yet? Just dropped, yes?
dropped 12 ish hours ago
Yeah its pretty good
openai
I rrally like their mcp native servers
Another name for this is "interleaved thinking" which is a thing for some models already, like Kimi K2, Claude, Minimax, Gemini and GPT models etc.
Are you sure? Iam pretth sure it only goes for agentic tools
And as far as I know gemini does not think inbetween calls
At least gemini cli
It does
That isnt the same
š„
its the basis for it
o3 picked up the rest lol
5 and beyond greatly improved it
in any case glm showcased it because it helps their models allot
We are talking about tool calls and etc inside of a agentic worflow like claude code for example
How do i get comet on my iOS again? I have perplexity Pro
Comet isn't on iOS yet
yeah all models beyond o3 are capable of it
Aah then thats why i canāt find it on AppStore
Is there any aprox. ETA on iOS release?
Are you talking about codex
Hopefully before end of year, no official ETA
8 days left of the year lol
Aaah cool cool, lets hope so then haha
Because u cant call tools like that inside of the web uu
any model of openai beyond o3 that can use tools during its reasoning
eh it can lol
when it needs to create files or unzip them before working etc it will use its own environment and cli tools to prepare them for work
We arent talking about the same reasoning
MiMo
Wow, it's so smart.
you mean the preserved thinking that also keeps tool usage in its context right?
Very Very smart, Genius model
wat
agi level reasoning
i c
hello
Now I want some horchata.
wats that?
it's a rice based drink. delicious. add rum and you have Rum Chata. LOL
oh wow lol
ive been prompting for 4 days straight for it to kind a break in the end, 25+ sessions, recalls and rebuild from recalls
how accurate is perp realistically when it comes to coding something extremely complex? (for context, im having it conceptually understand the system im trying to build, which is a very advanced pipeline automation system thats extremely flexible and can micromanage tasks, with maximum integration with whatever softwares needed, all dedicated to different personas which focus on multitude of things etc. kinda my own jarvis)
Perplexity is not made for coding, I'd suggest you look into Windsurf/Cursor for something this specific
are their deep reasonings solid for indepth contextual/conceptualization? or would i have to prompt for prompts to feed them better
It depends on what exactly you are looking for, it's best to use the reasoning models like Sonnet 4.5, Kimi K2, for in depth results
okay thank you ill try play w windsurf/cursor
use codex
would you guys be down to work together for this build? and we open source it as a community thing
its gonna be insanely cracked out sovereign localized nd sexy
Baubleee
seems like we are overdue...
welcome back
here it is
Bauble
New!! 14 to go!
can't even see it
Same
yep. just a poo
@dry kestrel woah pplx watermark for screenshots
Same
wht the
A Christmas Poo appeared!
dangnabbit I missed it
Feel free, it's just a š© not bauble
wats cool
Send me one.
Wow
Fuck no
Hey guys it actually worked
Yes
Interesting how the branding only shows when the model used sources
What worked?
Oh itās u
something from yesterday
Heās the guy yesterday who have the school mark down grades for charging laptop at school
OOOOOH!
sorry, didn't remember your username.
its fine
What did you end up doing that worked?
oh yeah perplexity
Noooo
another poo....thanks, Bauble God
charged my frikin thing overnight
Told u
I need poo baubles
Thereās nothing need to worry about
Nice!
Part II
Jesus born and Jesus Risen
#35 should be this one
Cooking
wat
Ultimate power
wat
Haha just joking
U know what
Its pplx event
2nd times the charm
ALL HAIL THE BAUBLES!!!!
Bubble
Oh ye 3 months of free pro
This will stack on ur current plan
Check pinned message
No like some people ahve the pro next to the username
I have not yet used Easter 3month
Excuse me, why the button will be expired in 5 minutes in the event-commands channel now?
Yo
@twin grail how to get to pinned
Started a bit ago. No idea. Keep the channel clean maybe? š¤·āāļø
I see! I used to check the rank from others command before.
I have to send the command by myself now
me too!
they paginated the collection view too to make it smaller in the chat stream.
It's probably an efficiency thing
opus 4.5 is a fucking monster bro
on Claude platform?
what did that involve?
the same thing but with easter eggs
The second season of Fallout started.
I wish they dropped the whole season. I hate to wait between episodes
It's a series based on the Fallout video game
but then you have to avoid spoilers for the whole season. haha
okay that doesn't help me lol...dunno much about the game either lol
easy enough...or are u talking about other people talking about it lol
yes...other people. and then I feel left out when I have to put my fingers in my ears. š¤£
ah that's the issue then lol
Anyone watching Pluribus?
1997 game?
It's interesting
Fallout? It's a whole game franchise.
It is. Cool concept.
In a world overtaken by a mysterious wave of forced happiness, Carol Sturka, one of the immune few, must uncover what's really going on - and save humanity from its artificial bliss.
waat
sounds AI generated
š
The last episode was, interesting
do i plug my ears?
I think the last one for the season is this week?
Yes
Always great to get a good sci-fi series (and not have it canceled right away)
even better since it's by vince gilligan
Breaking Bad, Better Call Saul...
The actress that plays Carol Sturka was in Better Call Saul too
you guys into The Boys?
She is good. annoying...but good.
everyone keeps telling me to watch that
Not into The Boys.
the lawer
One bauble and I go sleep
hm?
I don't have a TV, currently. My last one S95D is melted
who watches shows on TV these days š
HDR10+ lovers, no one mobile can't display real dynamic HDR with proper tones and mapping
laptop, pc, tablet, phone, or 75" tv. Why discriminate? I watch everywhere.
Did that start yet?
oh! I'll have to add it to my watchlist.
previews looked good. but for some reason I thought it was coming out in Jan.
you like the previous ones>?
yeah
Perplexity Events is typing...
lol
how much typing is needed? just drop it!
wat
and here I thought
we were having a nice conversation
haha
maybe if you got Max
bauble interruptus...
so you like Rum?
sure.
@green crystal https://fxtwitter.com/bycloudai/status/2002347892535930934
for more context
ļøļø
ļøļøon OpenAI's MRCR benchmark, curated at contextarena.ai by @DillonUzar , Gemini 3 flash achieved 90% acc @ 1 million ctx
ļøļø
ļøļøthis performance is SoTA across all models, most SoTA models cant even go past 256k ctx
ļøļø
ļøļøat this length, you cant be using standard attention, it'll perform bad anyways, and ofc it'll be very expensive. (Gemini 3 flash is $0.5 in $3 out)
ļøļø
ļøļøSome sort of efficient attention is implemented, so thats why the price is hitting the same level as a linear/sparse attention model
ļøļø
ļøļøBUT linear attention (hybrid) is only good at long ctx bench, and suck at knowledge task. G3F is great at knowledge, even #3 on Artificial Analysis Index.
ļøļø
ļøļøthey cant be using any SSM/mamba variants hybrid (at least w/ standard attn) either as those suck at long ctx. Same as sparse attention, as you can see from DeepSeek 3.2's DSA
ļøļø
ļøļøSo what black magic did they do?
ļøļø
ļøļøguess i'llā¦
Yes it will be upgraded with rl to surpass flash.
ah that makes sense
@shut dagger this you?
i hope it can improve
in many ways, it sorta regressed compared to 2.5 Pro for me
I wonder why Grok got delayed
Even 4.1 is still in beta.
B a u b l e
@wise nova is tired and wants a bedtime bauble.....
One can only speculate, I suppose, unless there's someone from xAI here who would like to share.
b a u b l e
Yeah, I want to go sleep
The eggs have disappeared.
Rare & a New one! 13 to go!
another poo
I mean the baublers
finally I gotit
no idea what it looks like...but got it! haha
Who was this?
I can use it as manure i guess
wats that for
Cold
any good?
10 x 4 months free google ai pro š
how
China #1
wow
google ai pro userās referral code
That's a pretty good deal, I didn't know they offered 4 months free on referral
only in china?
Checking your list twice? Google AI Pro members can give the gift of higher access to our AI features and models in two ways:
1) Share the love: You can share your Google AI Pro plan with up to 5 peopleĀ at no extra cost.
2) Gift a trial: Send a friend an extended 4-month trial
Gemini is not available in China
then how is he getting that offer?
Why don't they allow people in China to use Gemini?
well most Chinese people dont use google products already
Users bio say "HK", they have different regulations there
Google product
You can review the allowed regions for Gemini product in their docs
ahh okay thanks :)
The Gemini web app is currently available in more than 40 languages and over 230 countries and territories.Ā Weāll gradually expand to more countries and territories in a way that's consistent w
wow
Its freaking 5:47
Cant sleep
All sleep servers are full
Bauble fever?
sleep servers?
same thing ig?
man perplexxity annoys me so much sometimes, why is it not able to access my mails š
it directly denies that it cant check my mails
Did you add the connector?
guys what is this event ?
lol
I wonder if people had like a 24 or 48 hour trial of Max
if more people would use it
would buy it i mean
i havent seen anyone hear talk about max vs pro output
i guess you don't want people abusing Max too
so it cant be that big of a difference
that's cos hardly anyone uses Max
or have actually compared
it's more about limits
rather than output quality
I doubt the output quality is different
The point of a trial is to introduce the services and features. It doesn't make sense for Max, since Max is basically about removing limits and just adds one more model.
yea used to be 2 more models
A trial lets people answer the question 'is this for me'
well apparently can access GPT-5 Pro too
you get access to better models, therefore better output
berd
via complexity
he was clearly talking about for the same model
hello
please. i need a voice of reason
its been a month on my ticket. i know you dont work here
oh wow
i just need someone to listen or im talking to the void at this point
I'm honestly not too sure, perhaps try opening a new ticket starting with requesting for human support?
berd can you tell me when we get the rewards from this thing?
is it just a month for each one?
Till dec 25
top 10 gets 3 months
It's like a purple lava lamp. I'm surprised there isn't an ice lava one, for ... ice lava man
learn more here #š°āserver-news message
I got till top 10 in under 2 days
Most people will probably be offline christmas eve+day... I know I will be.
So others might catch up
I'll probably stop collecting at 45
about 60 hours left right?
Ends at <t:1766702400:t> on December 25, 2025
@Pirat_Nation >require ID for porn
ļøļø>database gets immediately hacked
ļøļø
ļøļøWho could've seen this coming??
**ā¤ļø 115āšļø 1.5Kā**
Anyone wanna try their luck for lifetime vpn giveaway?
Tailscale?
NordVPN?
what VPN
yes
fow what
I guess the promotion makes a lot of sense for them, seems like it's designed to try to incentivize social media outreach, and the brand isn't highly recognized.
wonder if the 3 months has expiry date to use?
I'm pretty sure using it on top of an existing subscription will extend the current subscription, so shouldn't have to worry if you're plan has a lot of time on it left.
yea was more asking expiry to use
like let's say u don't apply it for 6 months
would it stlil be valid
right?
any support team here to my payment problem solved
not really in this channel
i think Kesku can help tho?
looks like a reddit poster has said Max is better than Pro
for the same model
like better quality output
Well that's official then š¤£
100% lol
"The pro definitely feel more shallow now, in the past 3 month its may take 5 whole minutes to research any topic but its would be so information dense im pretty sure my phone lagged lmao, now? Yeah its only 2 minute and then the result is fast as hell
Max meanwhile, its still take long as time, and its make my phone lagged but its very information dense"
the problem is that
I don't think people actually know how to conduct a fair test
and many (most?) assume that just cos you do it once
it means that "one time" is always the case
that's just not how AI works
especially when the prompt requires search(ing) web
please send all the details to our support@perplexity.ai team, they will check and help you with this
does anyone know how those shared around benchmarks are done?
like is it just one run?
or SPRT level runs
wait not SPRT
but I guess "fixed games"
like fixed number of runs with error margins
or is it just "test position" testing?
And what are they using for testing. Is it an objective outcome or are they not even evaluating the output and only looking at speed?
He talked about information being dense. It sounded like research. Is it accurate? Written well? Cited properly?
i think they need to start putting error margins
oh that one...no idea
that's the thing....it's always a vague claim
Yep.
Instead of min maxing, everyone should use whatever gives them a result they like. Not everyone will agree.
perhaps
Sam is ai
Citation in pplx has been so buggy for me
Its answering questions according to Netherland website yet the citation are all about UK
now it's like 1/4
or 1/3
if there are 10
u'd have to click 10x
to look through them
rather than just a big list
that u can pick to click/tap
surprised they made that change tbh
.
That's not how pplx works
it's demonstrable that's exactly how it works
the citations of a response
it should to be when u hover/tap one
Jen
It you to the website
Got it
lol
And it was a new one too. 12 to go
New to me. I can sleep happy. š“
the citations may be different with app vs browser etc?
Browser was the same thing
oh wow it's different
in Chromium Windows
and Chromium android
web
I actually now prefer android way lol
when tap it, it displays all the citation
oh wow
I think complexity may already do that
not really
even Complexity changes imo
but at least u can see
what model
without having to check/tap it
yes it will still bug out
i've seen GPT-5.2 Thinking go to GPT-5.2 non-thinking
even with Complexity
but with Complexity, u can easily see it
and just change it back
yea it just uses the rest settings thing
Hi
lol
Bauble lunatic
that's Jen imo
no u
https://www.perplexity.ai/page/create?
HOw to access perplexity create new page endpoint?
Max might make it easier
alredy pro plan
0.001 $ per ad
š„²
Oh no! My identity is no longer hidden š
@drifting shadow Hi
Hello
why do i have error 'Gemini 3 Pro was inapplicable or unavailable.'?
Hello
hello?
Retry
just get Max imo
Mwahahaa
Peasants waiting all their life
facts are facts
Just to not get first
Max would fix it
Bro stop yapping
no u
At least tell a story
out of context
for example
that last interaction
is a genuine fix
your implication is that it's done out of advertising or spam or whatever
but you haven't countered
logically
that it's not a genuine fix to his issue
My implication is that you are yapping for baubles 
let's say that was true
what's wrong with that?
the literal theme is to keep chat going
this is literally the quote
"The Christmas Spirit is getting ready to sparkle in our ā š¬āgeneral chat from now until...dropping shiny Christmas baubles for you to collect as long as you keep the chat lively and active. "
,
l
Does anyone here use the Comet browser daily?
likely
seems good
wow
Thanks For sharing the detailed video Mares
very useful eh
Has anyone ever gotten any personal information leaked while using Comet as much as I have heard online?
dun Think so
Lol... I'm not much about video. I'd skim through text of do a trial and error
Definitely. I give it to comet to summarise for me and will ask them to make bullltins at last
yea
So far I'm using em like a browser with chat-ability. Never gave them access to anything more that a email ID.
e-mail ID?
yes
wats that
we use some things like Gmail.com , outlook.com, Yahoo.com, Yandex.com like that
okay
wats that
An e-mail ID
really?
Imagine ssj102@max.com
wow
Or ssj102@perplexity.max
Max?
seems you the one chatting lol
submail was anotehr good one
now it's gone
Flash better?
Is the new Xiaomi LLM coming to perplexity? Would be good for the speed. It really sucks meanwhile... https://ai-mimo.com/
I want Mimo!
What is that disco ball up here ?
Benchmark?
It's definitely not coming to pplx it's only 7B
š
Pplx team cares about model popularity so it's a clear no
Hewwo
I am new here
Fuck these Benchmarks. Try your use cases
wat
Some benchmarks can determine the use case
Thanks miC
what if exception tho?
@green crystal I have a feeling that the system instructions are different for different levels of reasoning.
I was talking about api.
Perplexity should consider glm/qwen models.
they did Kimi?
Yeah but kimi doesn't support vision.
Yes
Nope
weird
Glm 4.7
They really are missing out on glm
Glm 4.7 is likely the best model for agentic uses such as controlling the browser
How do I open this menu
I can't.
Looks chatgpt plus level.
I don't want to have to use a laptop/mouse just for this.
Try the desktop mode
Use desktop mode
I think its similar to gemini app
It effects your thread
As in the first query
It'll probably need mouse hover.
No i did it without pc
It doesn't appear in new chat either.
Type something
Enable the advanced search after you have written your query
Also
This isnt the vision model yet
It's glm 4.7
Their latest image model is glm 4.6v
To get a good experience they do
Otherwise reasoning would be roughly concluded lol
Api I think it's still very changeable
Since there's no sys prompt there
The reasoning effort is a more internal basic instruction I think
steals all the kills
Hi
Hi, so currently which model are you guys using for searching or writing?
No, specific model for search, just Perplexity. I don't use AI to write directly, but I typically use Claude when I want to review what I've wrote.
Gemini flash for search Gemini pro for research
Gemini flash thinking for searching because i like fast and high quality responses.
My man 
what are you writing?
Told kimi to follow usdv values but it keeps sticking to icmr values.
Typical failure in instruction following.
What is that?
I hope he didnāt use Comet to complete the course
ayo guys, do anyone know why don't models reliably write code longer than 600 lines?
Entry level data center role on networking, virtualization, storage area networking, unified computing, Cisco ACI.

