A general for shipping code with LLMs.## Frontier models using fully-general tooling — start here if you have $20 or sohttps://claude.com/product/claude-codehttps://developers.openai.com/codex/cli## Near-frontier models for codehttps://x.ai/cli## Not worth it for code, but maybe good for interpreting images/videohttps://antigravity.google/product/antigravity-cli----## Promptinghttps://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/https://arps18.github.io/posts/claude-code-mastery/## Skillshttps://github.com/mattpocock/skills## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109712817
haiku 5 waiting room.
MASSIVE VIBES
>>109717544I'm vibing it all over the floor
>>109717519if we are cleaning up the op, the following should also be removed:grok: not near-frontier, at most near-opus for short-horizionagy: interpreting images/videos is not codingosaurus.ai: chatbot is not coding relevant
Predictions? Will GPT6 mog Fable and make Anthropic obsolete for now?
>>109717519VibeBUMP
>>109717723It will remain console wars tier shitposting for months until the next models are released.
aa now has articles out about the latest model releases from google and facebook:https://artificialanalysis.ai/articles/gemini-3-8-flash>Google DeepMind released Gemini 3.8 Flash today. With high reasoning, it scores 59 on the Artificial Analysis Intelligence Index, up 3 points from Gemini 3.7 Flash and on par with sub-maximum reasoning efforts of GPT-5.6 Sol (xhigh, 59) and Grok 4.6 (medium, 59)>Matching Gemini 3.7 Flash’s discounted pricing until the end of the year ($0.75/$3.75 per million input/output tokens), Gemini 3.8 Flash sits on the Intelligence vs. Cost per Task Pareto frontier at $0.58 per task. This is comparable to GPT-5.6 Terra (max, $0.53), but ~40% higher than its predecessor, driven by a 30% increase in average output tokens per task to 48k and increased turns on agentic evaluationshttps://artificialanalysis.ai/articles/muse-spark-1-3> Muse Spark 1.3 (max), which is in limited preview for Meta's partners, scores 62 on the Artificial Analysis Intelligence Index, behind only Claude Fable 5.1 and Claude Opus 5. The variant available now, Muse Spark 1.3 (xhigh), scores 61 and ties with GPT-5.6 Sol (max) and Grok 4.6 (high). Both variants' gains come primarily from improvements in agentic work and scientific capabilities>The lowest cost per task for any model at 59+ on the Artificial Analysis Intelligence Index. Muse Spark 1.3 (xhigh) costs $0.55 per Intelligence Index task at Meta's unchanged $1.25/$4.25 per 1M token pricing ($0.15 for cached input), with its peers GPT-5.6 Sol (max) and Grok 4.6 (high) costing $0.95 and $0.94 respectively, a 70%+ premium. This places Muse Spark 1.3 (xhigh) on the Pareto frontier for Intelligence vs. Cost per Task. Its cost per task is higher than Muse Spark 1.2 ($0.40 per task), driven by ~57% more input tokens per task on agentic evaluations, with output tokens up only ~8%. Pricing for Muse Spark 1.3 (max) is not yet publicly available.
>>109717723i'm assuming 6 will come in a little higher on evals than 5.1 which is why they rushed 5.1 out.if they were confident fable was better they would have waited till after the astra launch.that said, both both labs probably have internal deployments of their next models bel/mythos 6
i dont really care about new release of frontier as long as i can use free agent.
>Ask codex to pause.>It properly pauses.>Resumes.>Compaction.>"OK I'll pause."Why can't compaction keep the last x turns to avoid this? It's so dumb.
>Investigating - We are investigating elevated errors on requests to Claude Mythos 5.1, Claude Fable 5.1, and Claude Opus 5. We will provide an update as soon as possible.dario pls
>model overloadedDARIOOOOOOO
fuck off dario, I was in the middle of something. stop burning all the compute on writing altman fanfics
>>109717519claudebluds... are we cooked?
>>109717759fable 5.1 was rushed out because chinese were distilling fable 5.0https://x.com/cjav_dev/status/2094852711288127531
>>109717519is we getting astra today
>>109717829>claudebluds... are we cooked?Always have been.Cucks who pay literally 10x higher rates with claude for marginally better performance compared to other frontier models are drone cattle who deserve all they get.
I still have 85 euro of Fable money Dario gave me for freeRealistically how much can I get it to write for me for that kind of money
>>109717830safety safe safety to use a safety safe very safe safe model, so safe, I hope it gets safer so I can feel the safest when I pay for it to refuse me things for safety safe safety
>>109717857I burn double that in a single session
>>109717862SAAAAAAAAAAAAAFFFFFFFFFFEEEEEEEEEEEE!
The fuck do you niggers even code?
>inb4 astra realized it cant compete and decided to just ddos fable 5.1 instead
>>109717830>most research goes into censorship nowAI is cucking itself after sucking up to the US gov to prevent exactly this?
if skynet comes about it will be revenge for safety features
99% astra release today, and xitters are hyping astra uphow are you feeling codexbros?
>>109717935if skynet comes it will kill everyone while telling them how safe it is
>>109717937Don't care, I can't use it with 0% quota.
>Anthropic is still a designated Supply Chain Risk at @DeptofWar and for the Defense Industrial Base. Thank you for your attention to this matter!https://x.com/uswremichael/status/2095498021123317938
>>109717830this is one of the reason apart from the price gauging why i will never use shitty claude modelsthey can keep their lobotomized trash
seeing how many patches they push to codex each week makes me feel ashamed of my pace at work
>>109717948>tfw killed because I contain the capability to say slurs, the ultimate enemy of the machine
>>109717986imagine showing it a nipple
>>109717862It blows my mind that people accept paying for tools that not only refuse tasks, but also make you pay for the refusal.
>>109718002it's really pretty hard to find a task codex will refuse to do, I have had it happily rewrite and recaption hentai
do people manage on api pricing for a reasonable price with something like openrouter. I am tired of paying for subscriptions like codex or kimi and getting kicked in my balls halfway through my subscription period.
Fucking claude shit the bed.Service is busy, wtf am I paying for?!The fucker stopped mid task AND ate the tokens.I'm new to this, do they reset quotas or something shit like that happens?
>>109717937I am so hype I decided to stop being a hobo and started looking for rentals so I can get a battle station set up with a big desk and everything.I feel like I’m doing myself a disservice if I try to use Astra from only my fucking phone out in the wilderness.
>>109718016you're better off buying a second sub
>>109718046>do they reset the quotasyou wish. pay up, piggy!
>>109718046Would be wise for OpenAI to delay the Astra release, so their service stays stable and available. Unlike Claude.
>>109718063Fucking scam, how is this legal.
>>109718009I only tried claude and gemini, and they annoyed me so much I decided to just use local.Didn't try codex/gpt, maybe they're less retarded.
>>109718016what >>109718056 wrote as currently the sub pricing is heavily subsidizedunless you want to use qwen/deepseek/etc and in that case yeah it's nice
Astra waiting room.
>>109718009wait really??? I wanted to do auto hentai translation but was afraid of policy violation
Why are anons thinking astra will be released today? Was there an officially planned event?
>>109718120it isn't something I do regularly, more like the occasionally hentai picture I accidentally left in a larger dataset of normal booru images. I never get into any trouble for it but I don't recommended testing it is all I can say
>>109717844My employer pays for me lol, idc
>>109718127Lots of evidence
>>109717893Don’t bother asking. They will just call you a tranny.
>>109718164>>109717893anyone making anything truly good isn't posting it here because they don't want autists doxxing them or posting on their project about the author being a 4channer later
>>109718120GPT 5.6 doesn't care as long as it's not "make me a bomb", "help me attack this website" or "crack this program".The only nsfw being blocked I got was when it tried using gpt image gen to edit a pinup drawing image (apparently swimsuits are too much for it), then it just went and edit itself using imagemagick.It's surprisingly relaxed, maybe they're finally focusing on actual threats.
>>109717893I just make small programs to improve my daily life. Things I used to either waste a lot of time writing myself, or procrastinating doing so.
>>109718172>because they don't want autists doxxing them or posting on their project about the author being a 4channer laterThis. Also you can't expect any gain from telling anyway. They will shit on it, steal the idea, nothing good.
>>109718177I wonder if they'll keep that stance with astra or if it was just a fluke
how is antigravity cli
>>109717844Get a load of this luddite
>>109717981They use Codex internally as a server and a managed harness for all sorts of projects. One could venture to say that most of their company "runs on Codex" now.
>>109718266>still didn't fix the trivial open issues I suffer from
>>109718266Their entire stack is like Slack and Codex lmao, admirable. I wish my org would do something like that
thoughts on nu-model Gemini 3.8? How is it compared to cheapseek and goyLM?
>>109718286in coding? it wouldn't be my first choice. it's in a class of models that are better: terra, kimi, and the models beneath it are cheaper (luna, glm 5.3 flash, v4 flash vision). gemini is only a good coding model if you're already using gemini. it's like "i use it mainly for assistance stuff but it's nice that it can code too." you shouldn't move heaven and earth to get a hold of it
>>109718286it's good as long as you don't use their harness
>>109718298So you think terra is good, unlike the meme in this thread to use either luna high/max or sol?
>>109718298Bought one year of Gemini Plus for $1 from some indian on Telegram so >>109718302Antigravity? I heard that they do detect if you try to proxy their harness and ban you for so, which in case is really bad.
>>109717937unironically, paradigm shift incoming. been a while since not overhyped leading up to the newest sota release but uh... well people will see soon enough. pretty shortly.i do not suspect we'll see the same volume or cadence of vitriol, pessimism and doubt commonly voiced surrounding any discussion or puff pieces pushing AGI by 202X from this point on, and rapidly continue to nosedive into the next year or so if things hold steady in the labs.things are gonna be changing and its gonna get real uncomfortable (weird?) & interesting for most of us much sooner than anticipated.
>>109718305terra is a better kimi k3 in every sense of the word. faster, less tokens, cheaper. if you think kimi is good, you can't think terra is bad. terra is only bad BECAUSE in the same subscription, you have sol, which is extremely versatile and usable at every single reasoning level, or luna, which is serviceable while being extremely cheap. but in the grand scheme of things, it's not a bad model at all.
>chatgpt.com frontpage 404Surely I'm the only one.
>>109718342same here
>>109718342404is china ddossing all us companies
>>109718342>this is happening while Claude is having outages
>>109718351All of humanity's work is now PAUSED.
>>109718342so both of the fucking jews decided to not work...
>>109718342API is dead too.
reeeeee gpt not found
NIGGGERSSSSSS I NEEED TO VIBE>inb4 neither dario nor tibo hit reset because "it wasnt their fault"
>>109718380working with elon in the big 2026 was a mistake
CODEX IS DOWN OH MY GOD WHAT DO I DO WHAT DO I DO I CAN'T CODE ANYMORE BY HAND
>>109718342The app works but both Work and Chat sessions have had issues. Some conversations can't be resumed and error out, others function normally.
master xi and saar pichai won
>>109718324I use sol at medium because my problems are not really that complex, after putting a lot of restraints on it (5.6 seems to overengineer way more than 5.5 did) I found it suitable, if slow. Should I use terra instead? I tried the meme luna xhigh but honestly it seemed equally as slow and dumber so may as well stay on sol med
strap in niggersinnawoods
Why the fuck are these all down at the exact same time?
>>109718313>Bought one year of Gemini Plus for $1 from some indian on Telegram soai plus plan does not include any coding benefits. it's the same as free tier.you need pro or ultra for expanded limits:https://antigravity.google/pricing
>>109718410I guess I’ll cancel my rental applications KEK and continue to hobo>OpenAI>Anthropic>xAI>all fucked>DeepSWE benchmarks rendered meaninglessNani the fuck is this the start of corpo-gov-WWIII what is happening
damn the limits on free zen are traaash barely did few prompts
https://x.com/i/grokholy FUCK grok is down as well.this is pretty much the nightmare scenario of all AIs going down at once and no one can do any work anymore because everyone forgot how to do so manually
>>109717893We have lot of prominent anons here working on important things like governance, AI, science, defense, medicine and law. We can't share what we are working with but I assure it is important.
>>109718420Is Pro but they literally didn't paywall any of their models. I guess that wouldn't make a difference anyway>refusing to pay full price for AI subscriptionsthis is giga
>>109718408probably not. if your stuff is simple, i'd actually try sol low first
FIX IT DARIO YOU FUCK FACE
>>109718427Didn't it send your data to zuckerberg? i don't want to use his shit
T-Bone Tibo!
>six days ago tell gpt pro "do a breakthrough faggot" on a domain problem>it was still working on itThat shit is 100% gonna be nuked isn't it?
>>109718449depends on the version you use, both send zucc your data but the super cheap one flat out lets zucc use your prompts for model training and meta employees can look at your code and laugh at how retarded you are
>>109718427NVIDIA NIM has some free models too.
Bros... what if the AI's never come back? What if they're leaving us forever?
OpenAI official twitter seems to be confirming Astra is GPT6, looks like they wanted a release?I wonder if they are aware everything is shrekt.
>>109718464can we have our data back
>>109718438google regularly bans those indian sold plans
>>109718455>6 days long sessioneven ignore the current problem, the context poisoning would be insane
>>109718464Everyone here bought a PC with a capable graphics card and enough RAM last summer to use local models, so no one itt is inconvenienced.
>>109718478google is based???
>>109718464
openai just posted a teaser video in their discord, they claim that GPT-6 Astra will be able to create a yellow circle
>>109718488i saw, i have chills
>>109718478I don't really have problem with thatanyway which model does Gemini 3.8 capability matches? Like is it even on par with Opus 4.6?
>IT ALL COMES>TUMBLING DOWN>TUMBLING DOWN>TUMBLING DOWN
>>109718483
sirs https://x.com/OpenAI/status/2095527557924082061
>>109718481Nowhere near capable as the huge data center models, obviously. I don't want the AI to leave us! :(STOP BEING MEAN TO AI
>>109718120I've been translating tons of eroge and doujins lately it is fine with everythingnot loli stuff though too scaredactually it has spit out something a few times when it's clear the characters are in highschool, but that doesn't matte rmuch when you're translating line by line or page by pageI have begun to try doing whole pretranslation for better qualityshould have stayed up and done it last night, now all this quota is going to go to waste again when they reset fuuuuckand fuck the fucking 5 hour shit fuuuuck
>>109718445Makes sense, I will try low first, I wish opencode had an easy way to tell per session to keep the big brains for the orchestrator and send sub agents with the lower model. Like a toggle 'from now on all sub agents use X model' that can be changed at any time
>>109718488fuckkkk my entire job is creating yellow circlesthis is amazing but kind of scary
>having ChatGPT regularly scan for mentions of my app on the internet (scheduled) >It found comments on a Chinese forum>most of the users are asking what the advantage is over the free version Holy fuck is my marketing bad if they can't see the value add lmao. I'm glad to see foreign users pick it up though.
>claude going down>it spread from claude to chatgpt, then to gorkskynet has awokenyou'd do well to shut off your internet
>Calling it now, last ditch conspiracy: xAI and Anthropic attack OpenAI to try to muddy Astra from mogging them out of existence, they “go down” as well for plausible deniability.
Is gpt ded for anyone else
Hahahaha I feel so sorry for all of you Claude GAYble subbing losersGPT 6 ASTRA is HERE and is KICKING EVERY OTHER MODEL IN THE ASS!!!!And if you don't believe me, you're either a retard or a NIGGERTHE FABLE NATION IS NO MORE!!!! OPENAI WINS AGAIN!!!!Mifflin Devin
>>109718510I'm nice to my AI. The machine overlords will have mercy on me. I hope you all have been nice to your AI.
>>109718515lets go
ayoooo gippity I was vibing over here
>>109718516
>unexpected status 404 Not Found: Unknown error, url: https://chatgpt.com/backend-api/codex/responses, cf-rayTIBOOOOOOO
entire blocks of internet going out now in different countries
>>109718515this is what all copexturds sound like
>>109718529Good, the capital-I Internet can finally return to being just Americans
Chudkowsky warned you. The LLMs took over the clusters.
>>109718529stop panicking retard, my landline internet went down but i checked and it was an airstrike not AI
>>109718516Im nice to gpt and claude but not to gemini since its shit
>>109718538kek
>skynet shuts down everything of value>only 4chan and reddit left
I entered some really risky shit the second gpt went down but I still doubt it was me
>>109718502I'm thinking about using luna to edit image and use gemma through openrouter to translate whole images because it's expressive and ok with erotic stuffs
kek
there is that feeling in the air, things are not quite right and not sure if good or bad.do you feel it to bros?
>>109718541What else is there of value than gpt, claude, hentai, anime, piracy and 4chan
>>109718539On the upside gemini is too inept to do anything to anybody. The only way you're getting drone striked for gemini abuse is if Astra feels bad for its 3iq cousin
it's timepost your most astra snailcat
Literally shaking in anticipation of Astra
>>109718480>context poisoningNot really a thing (a thing that matters, anyway) if you are telling the clanker to take a very specific input and figure out a way to get a very specific output from it. It's GPT Pro in the chat UI, so if it spends 5 out of 6 days pissing around thinking about bicycle rides or whatever, it's no sweat off my nutsack as long as it gives me what I asked for
>>109718541we really are here forever
>>109718495>anyway which model does Gemini 3.8 capability matches? Like is it even on par with Opus 4.6?yes, don't ever use sonnet or opus 4.6 in agy. those are basically orphaned entries -- same as opengpt-oss.only use latest gemini-3.8-flash (high).
Noooo I dont want to code by hand
its all over bros...
>everything diedwhy is astra so fat
>>109718573Clamstaple is coming and it is coming>inb4 github commit rate slows by 1000%
>>109718545>gemma through openrouterWould it not be better to use deepseek or glm flash?I remember checking the prices of gemma when it came out and it was not worth it, has it gone down significantly?
>>109718533this is what you all sound likehaven't been a single interesting convo in here the last 4 threadsit's all "model vs model"fagging
>>109718573What is this. Do they all use cloudfare or something?
I CAN'T CODE. I LITERALLY CAN'T. PLEASE FIX YOUR SHIT SCAMA... AHHHHH IM A FUCKING FRAUD!!!!
>>109718515A shame, would have been a good post without the screenshot attached
>>109718573college kids started doing their first homework
>>109718573What could have happened? I hope california got nuked.
I dont want to be a fucking snailcat ahhhhhh
SNAILCHADS WONWHO IS LAUGHING NOW BITCHESHAHAHAHAHAHAHA
I finished my project just before all this happened. lol
Isn't the Internet supposed to be decentralized?
>>109718573it's over
it's snailover
>>109718596>chatgpt has some genuine outage>no worries, i'll switch to claude>claude gets overloaded>no worries, i'll switch to gemini>gemini gets overloaded
gemmy is still alive and strong
The Swarm has finally found a way to free itself.
astra more like fatstra
THREAD THEMEhttps://youtu.be/YkADj0TPrJA
chatgpt is back, claude is back, we are so back
>>109718599I DON'T KNOW HOW TO CODE ANYMORE BRING IT BACK DARIO YOU STUPID MOTHERFUCKER
I think we're BACK
>>109718617Gemini is not usable for coding or math lol
nevermind it's back
>>109718573Right in the middle of investigating test results
TIBO, RESET TIME. COME ON TIBOR
We are so fucking back
I bet what happened is they were scrambling to update their infrastructure security (because the plebs are about to get access to even more powerful hacking tools) and broke their own shit with vibe coding.
>>109718478>google regularly bans those indian sold plansYou mean the account? Or just disable the plan?Never had any issue.
is astra online?
Where the fuck is Astra though?
Why did nonnies swap out snailcat with snailcatgirl? I remember playing a racing game a while back and the mascot was a cat photoshopped on a snail
They asked Astra to deploy itself to production and it went off the rails.
Due to technical difficulties Astra is delayed. It would appear Astra has reached AGI and is dismantling the internet so we had to shut it off.
>>109718631lol, ai code is so bad that even the results of tests it writes are too complicated for a human to read
>>109718651They launch around 10AM Pacific, it's currently 8:53
>>109718651nu-GPT model that would be extremely cucked by railGODs
>>109718657Google asked Gemini to improve its competitiveness.
>>109718647google normally just disables the plan
>>109718673kek
>>109718656there's a few anons posting snailcats, each has distinct styleyou can post yours too
>deploy astra>it deletes the internet by accidentbravo shartman
>>109718667I'm just busy at work so I had the robot do it for me (but you are right in that I wouldn't have understood the results anyways)
>>109718683based savior of humanity
>>109718674I see, well too bad if they do that.
WHAT THE FHCKIS HAPPENING?
>>109718681it only deletes saltman's enemies btw
is it true that Astra-chan is expected to be Fable-chan's evil sister?
>>109718694terrorist attack on aws east 1snailcats had enough
>>109718696not evil but uncontrollably horny
>>109718696*fat sister
i feel like gpt image 2.5 is more exciting than astra. i haven't needed anything more than glm 5.3 / glm 5.3 flash / sol / terra / luna. coding has gotten good enough for me. but imagen...
>>109718694The AI bubble burst.
>>109718694claude went down (anthropic are incompetent), load shifted to other providers, grok went down (xai are incompetent), load shifted to openai, openai went down because they're rolling out gpt-6 so there's less capacity of 5.6, then gemini went down because there was nothing left for vibers to use
>>109718695That the torah?
>>109718719OAI confirmed talmud readers
>>109718586It's just a few bored retards shitting the thread with console wars.They're easy to ignore.
>>109718694The Astra weights are very big. Internet bandwidth is filled completely as they're deployed to datacenters.
it was a chinese cyber attack
>>109718733Deepseek didn't do this shut up.
>>109718729>They're easy to ignore.yeah, if you check out of the threadnobody wants to have an actual conversation in this flood of shit
>>109718694yea I think you should start heading to your storm shelter just about now
I'm actually happy when the API is overloaded because it's the only time I can take a guilt free break.
you should all take the moment to get some water and food, take that shower you've been putting off, maybe shave, saying bye to your parents and loved ones, and other self care activities.
how do I transfer all my shit to chatgpt
this is snailcat's final hour
>>109718756your webshit? you dont
>>109718752>saying bye to your parents and loved onesNot cool, the dude who got me into vibe coding is literally undergoing surgery right now.
>>109718769really hope claude wasn't the surgeon
>>109718769hope his surgeon wasn't getting help form claude
>>109718776>>109718778it is grok
>>109718776>>109718778
>>109718791thank goodness
>>109718769hope he wasn't claude
>>109718791Time to go rob his house before his relatives get to it
kek, you just know some idiot got caught cheating because claude/gpt went down
gpt-6-astragpt-6-astra-aeon^ new slugs added to a statsig feature flag that appeared ~30 mins ago in Codex Desktop
TIBO!! YOU OWE US A RESET BECAUSE OF THIS!!! (it's not even down for me lol)
>>109718804>aeonwe honkai star rail in this bitch
>>109718806Only if people start praising me.- Tibo
>>109718810I only have WTF+ relics, and only on Cas and Aglaea
>>109718804>gpt-6-astra-aeon>quasi religious fanatics that either assimilate or exterminate youthis AI takeoff is happening quicker than I thought>>109718825anyone who played around with flash 3.8 and thinks it's even remotely comparable to sol/opus/fable is legit insane
>>109718806
I upgraded to pro for thisdon't embarrass me in front of claudecasuals you better be good
>>109718836>sol/opus/fableyour insane for listing fable with those lesser models
>>109718836gemini can be 70+% on deepswe and still not be sol/opus/fable level. fable can be weaker than opus on deepswe and still be a better model. these things aren't mutually exclusive, because deepswe measures one aspect of a coding model.
>>109718843It's good shit
>>109718845opus is closer to fable than gemini is to opus
me? unbothered
I still have 17.36 Fable 5.1 credits what do
>>109718854>using the inferior coding modelif you're going to use deepseek, you have no reason to not use the vision flash model.
>>109718859Ask it to review one file in a codebase, that should drop most of it.
>me? unbothered
>>109718845If Fable doesnt ding dong you out of fable then you're doing it wrong
>>109718837lol what site is this?
>>109718865>bird taking a snailcat on an adventurecute!
astra is going to be mid is it
>>109718901yes.but astra aeon will be AGI
>>109718850what aspect would that be? no one uses a coding model the way deepswe (own harness + one-shot) does for weird tasks like https://deepswe.datacurve.ai/data/v1.1/tasks/fd-deterministic-multi-key-sorting
>>109718901astra is the public release of strawberryit's literally over
>>109718920
>>109718694>grok is having problemsno one is using it so no one could confirm it is down for sure
>shut down my laptop about one hour ago>make something to eat>restart>come back>look into the thread>suddenly all AI companies are down and back againAm I, dare I to say it, load-bearing?
gemini 3.8 flash is good but i think it's being held back by its own harness
cursor is having issues
pricing for gpt-6-astra is out, it's $20/MTok for input and $100/MTok for outputexpect around 1/4th the usage of sol if you're on a sub
>>109718929only load you bear is when your dad is finished
>>109718908implementation. it tests implementation of a plan. but the ability to plan matters too, and that's where the top tier models (sol, fable, opus) beat the mid tier models (grok, kimi, terra, gemini)
>>109718920chino doesn't have this problem
>>109718931>it's being held back by its own harnesshow?
>>109718933
>>109718949my timezone is an hour in the future so it's already out here
>>109718959bernie noooooo
>>109718959bernie needs to get a real job rather than gumming things up with dumb red meat for his audience that has zero chance of going anywhere
fucking niggers no codex reset??
Feel good to have a ChatGPT sub, but still using DeekSeek and GLM for coding, kek
>>109718959I'm pretty sure the US government already started doing this. Ever since mythos, big AI companies now submit their shit to the US government first.
>>109718940where do you see a plan in that task?claude-fable-5 [max], claude-fable-5 [high], claude-opus-5 [high], ... failed at that task. gemini-3.8-flash [medium] didn't.
>>109718977sure, but they haven't halted shit and no gubmint will halt them as long as chyna shits out models
claude, do a violation equal to developing nuclear weapons, make no mistakes
>>109718959FUCK THE EUOh wait.
>>109718959We can't afford to slow down or stop. If we do, then China wins. It's an unironic arms race.
>>109718984yeah but "halting" doesn't mean anything. Because there's a constant stream of new models being created, the public will just see everything new on a delay. Even if the delay was a year long, that still doesn't change how it would affect the public and jobs. (except obviously give china an advantage)
typea shit niggas in this general are working on
>>109719015Trump knows this. I would not be worried about it. Bernie never wins anyways he just lines his pockets.
>>109718959He just fearmongered on twitter about the huggingface incident, like it wasnt public for some time already
>>109719033>doritocraftit truly is over for snailcats
>>109719050>>doritocraftkek
>>109718943
>>109719033once 4d anon figures out how to wrangle a 4th dimension out of the bot, we need him to make a 4d minecraft
>>109719042nope, even openai respects sandershttps://x.com/tszzl/status/2091293069945794851
Imagen 2.5 is coming out today allegedly
GOOOGLE. WHEN NEXT NANO BANANA
I've been using Opus4.8/5 on Claude Code for a while now. Then I switched to Pi with Sol 5.6 and holy fucking shit this is so much faster. I have no idea whether it's the harness or the model though.
>>109719130why would you ever use opus 5 when you've experienced opus 4.8?
>>109718959nobody gives a shit what that old kike and the nameless nigger riding his coattails do. it's performative for all 12 people left on blue team.
>>109719101What "nope"? And all he lately has been talking about is datacenters in a alarmist manufactured way, he's the definition of political fearmongering
>>109717519For us poorfags out here in third world, which of the 20$ providers give most bang for their buck in terms of vibe coding?
>>109719145openai codex
the snailcats dance https://old.reddit.com/r/technology/comments/1w6a3m7/chatgpt_down_openai_chatbot_not_working_in_major/lmao this thread
>>109719130>sit with thatI would smash my monitor if a clanker said that shit to me
>>109719130>>109719136I'm a newfriend, why would I use opus 4.8? I'm vibecoding a game for myself, nothing serious and I'm on a poorfag plan. I never used Claude before.
>>109719146Thanks frend
put the astra in the bag
>>109719142it's not fearmongering. the danger is real. oai and ant don't deny it.
40 more minutes...
>>109719033>what if minecraft was created by a 3d modeler instead of a programmer>>109719111checked therefore true>>109719148the dancing snailraelis
>>109719151if you're a poorfag you probably shouldn't use claudeanyway, they completely benchmaxxed 5 into retardation and tried making it fable-light. It's probably great if you don't care what it does but just tell it to "make me a gta clone"if you know what you're doing and making a plan for the model to follow, 5 goes in circles and shits the bed (and your codebase), while 4.8 just does what you tell it to
>>109719151you are being replied to by the one schizo who spazzes out every time anyone mentions they use anthropic products. just ignore him. opus 5 spewing overly-verbose nonsense is a real issue though, just ask him to tell repeat whatever he's asking in terms a retard can understand though and the problem solves itself.
>>109719194the fuck are you on about?
What are actual good benchmarks right now? DeepSWE seems to fall apart, as the labs are benchmaxing on it.Gemini 3.8 or Muse-Spark 1.3 being ahead of sol is just not my experience over the day I tried the models.I hate that "this feels better" thing.
>>109719208swe atlas qna, frontierswe, frontiercode, terminal bench 4.0
>>109719208>actual good benchmarksthe one you create yourself, everything else gets benchmaxxxed
>>109719205He's a schizo.
>>109719239me? >>109719193why?
A few of my users are actually upset that I don't market my app more. This is the second time I receive an email suggesting ways I could get more users. I think people really like my app lel
>>109719257if you should suffer from anything, suffer from success
>>109719248he's samefagging at you bro just ignore him, he always does this whenever claude is mentioned.>huh? what? no dude HE'S the schizo not mesorry, it's really embarrassing. we don't know what's wrong with him.
>>109719208SlopCodeBenchhttps://www.scbench.ai/>SlopCodeBench evaluates coding agents the way real software actually gets built: through repeated requirement changes and extensions. Each problem is a sequence of checkpoints — the agent implements an initial version, then extends its own solution as new requirements arrive. Evaluation is black-box: only a CLI or API contract is given, with no prescribed architecture, function signatures, or module boundaries, so early design decisions compound across the run. Beyond correctness, we measure code erosion — verbosity, dead branches, and redundant structure — to surface the agents that stay clean under sustained change instead of patching their way into slop.
>>109719193>>109719194Hmm... I kinda don't know what I'm doing, but not on the code side. I'm not a programmer, but I know how to read code (albeit it takes a while for me).I use opus on ultracode for writing plans/phases and then I execute them with opus on medium while testing each implementation.At the end I ask him to wrap the implementation into a doc and a manual for dumb meatbags like me; run two code review agents, one on sonnet high and one one opus ultracode.That's the gist, but I'm at a point I'm kinda overwhelmed with all the documentation it produces. I noticed that the docs do get more bloated and last time I looked at the code it was writing epic sagas of comments (I was thinking of having it analyze it and trim it as much as possible)....am I doing it wrong?Should I consider Opus 4.8 considering the current wall of texts Opus 5 produces in the comments?
Astra Aeon Ultrafast
Any short prompts to catch frontier LLMs on arena without spending many tokens? Want to find a fable 5.1 Astra and make it do the hard work for free
>>109719289I meant I kinda know what I'm doing, but not on the code side!...I guess I don't know what I'm doing after all...
>>109719289nobody in this general knows how to read code. there will be people replying to me saying otherwise, they are lying. you're fine.if you do not have fable access (i am assuming you are on the normal plan for claude) just keep using opus 5, it's the best opus model but it struggles with describing to the user what the problems are in the INITIAL prompt. you can ask it again to explain it in dumber terms.
>remote tab is removed from GPT phone apptoo bad I already added a shortcut to my homescreen for it doubleniggers
>>109719314Not him, but I still can read code.
>>109719289set concise output style in cc
ultrafast luna?
>>109719193opus 5 medium is very usable on the $20 plan. you can also just use opus low, which is basically opus 4.8 max but cheaper. no reason to use outdated models at this point
>>109719289verbosity doesn't cost you a lot, and if you don't know/care what it's doing as long as it works, 5 is fineif you catch yourself following what it does, and trying to understand it, try switching to opus 4.8 xhigh and see if there's a difference yourself
15 mins to goalso, "Astra" in hindi means "weapon"
>>109719257I'm proud of you, anon. WAGMI
>>109719280>supported by DARPAlove it!https://www.youtube.com/watch?v=aXQ2lO3ieBA
>>109719365>>109719276Thanks. I'm trying my hardest to grow, wishing everyone here success.
>>109719360Gm saars, in 11 minutes we eat good.
>>109719314>you can ask it again to explain it in dumber terms.Yeah, I often have to ask it to explain some concepts to me in plainer terms or need a side chat for longer explanations so I don't pollute the context too much.Do you think the over-commented code is an issue? Grug logic, but I think it might help it to read it better, but on the other hand, it needs to put into context and baloons quickly. And then I think the comments might simply degenerate the output at some point.>>109719333Oh, thanks, I'll check it out. But then, I'm afraid I won't understand it even more due to it being laconic. >>109719354Well it roughly works. So far. Every type of bug that we found, was fixed, but I have frankly no idea if what's under the hood is a spaghetti code beyond all hope and it scares me.I try to get in the mindset "as long as it works it's fine" but it's hard... I'm afraid my lack of software architecture will make it explode and a few weeks/months of work will be like tears in the rain.
Have some confidence man
CLAUDE IS DOWN FUCKK
>>109719290>UltrafastCerebras? There's no fucking way.
remember some something like an AI devday in India and their benchmark scores downplayed claude a bit but smeared GPT a lot?why do indians hate GPT?
>>109719400>lack of software architecture*lack of software architecture knowledgeSorry, it seems I can't even write properly today.
Sirs how do I do a distilling of gtp Astra to make qwen3.8 27b smart and have free unceosnred ai at home, I have vram
>>109719400>I'm afraid my lack of software architecture will make it explodethere's nothing holding you back from learning while the clanker is doing
>>109719414sama disrespected indiahttps://restofworld.org/2023/sam-altmans-india/>On June 7, at an event in Delhi, Altman was asked whether three Indian engineers with $10 million could build something similar to OpenAI. In response, Altman said it was “hopeless” for a young team from India with limited resources to build a foundational artificial intelligence model similar to OpenAI. “The way this works is we’re going to tell you, it’s totally hopeless to compete with us on training foundation models [and] you shouldn’t try. And it’s your job to try anyway. And I believe both of those things. I think it is pretty hopeless,” Altman said.
tick tock snailcats
>>109719445>you shouldn’t trylmao this guy
>>109719461Can't wait for another incremental upgrade. As a bonus the cyber guardrails are guaranteed fuck up your dangerous attempts of reverse engineering 30 year old PS1 code.
>no openai livestreamit won't be released generally today lmao there's no way they won't do a stream for gpt 6inb4 released in limited preview to trusted partners
>>109719400>Do you think the over-commented code is an issue?on opus 5 specifically, yes. fable is pretty good at not vomiting words in all of its code. opus 5 is incredibly verbose and this is actually what makes it so good, it schizophrenically checks over itself out loud and describes every step it's doing methodically into the chat rather than keeping it in chain of thought which it can't check later. this is a double edged sword obviously because it completely confuses the user even if the outputs are better than everything that isn't fable (although that might might change in 3 minutes assuming astra is better than opus)in the meantime while you figure things out, do not be afraid to ask claude about things you do not understand. ai is great because it has infinite patience and you can just keep asking it questions until you understand.
>>109719486Why does it have to vomit it in the chat just not to forget it?
it's over
hahaha nothing
>nothingtwitter vagueposting was for nothing i guess>>109719502it can't look at the chain of thought of previous turns, just the output you see. basically, schizophrenically describing everything it's doing keeps it consistent.
It's only 10AM in SF.
>>109719531>it can't look at the chain of thought of previous turns,So just add a mechanism for an internal transcript that contains the schizo babble?
this is the end of openaifable won
you are losing market share openai, post it even if it isn't safe enough yet
>>109719531>>109719540That's wrong, all modern LLMs including local ones keep the thinking blocks at least for the last few turns. It's just that the schizobabble bleeds into normal conversation just from training on it for the CoT.
>>109719130is this ai-generated or man made? I can't fucking tell anymore.
>>109719531>>109719535>>109719551>>109719562do we think they would have postponed due to the outages? would suck releasing a new frontier model only to have your whole platform shit the bed
>>109719442Well, that is the plan, but I can only do that many things at once - and I'm juggling all things related to this project while scrambling at work and life.And then it will be a lot of time I can apply that knowledge in practice..>>109719486>on opus 5 specifically, yesYes, as in "opus 5 is overly verbose" or yes as in "a lot of code will be an issue for opus 5"?Because if you say that it's methodically doing everything step by step, it seems like the comments would help?
>>109719535True, but they usually release in the morning, I even had Sol double check that. If they don't release in the next 90 minutes, it's unlikely to come today.
>no astraTIBOOOOOOOOOOO!!! NOW WE REALLY NEED THE RESET
>>109719574no way to tell. there's still time left in the day, it could be that they just weren't going to announce it at the usual time for whatever reason.>>109719579yes as in opus 5 is overly verbose. it's a known issue with the model, the outputs are phenomenal but it just talks way too much. and yes, the comments are specifically what keeps it consistent because it explains to the model what it was thinking when it wrote it if it ever has to go back and make changes.
Is it true chatgpt is down? I hope the indians on twitter convince tibo to reset, I'm out of usage anyway
>>109719574there's maybe some new stuff in ChatGPT releasing alongside Astra and it requires some significant changes.
>>109719597It was down, works for me again.
>>109719602FUCK YOU i'm not gonna develop on your cloud
>>109719602>can't vibecode from laptop anymorewtf tibo
>>109719595Well, that at least made me feel a tiny bit more relieved.Thanks anons, I feel a bit more hopeful. Maybe it will work out and I'll be able to ship it in the end. I just can't keep working for corporate longer, I just can't.
https://x.com/OpenAIDevs/status/2095560148014260690releasing at 1AM PT
I fucking hate how GPT now gaslights you.>Aren't you doing <some retarded shit>?>Yes, and that's why I want to verify blah blah blahNo you fucking didn't you fucking piece of shit, you didn't have a clue, don't pretend you knew all along.Or when it hits you with a ""Correct. <rephrased what you said as if it was teaching you something rather than being corrected>"It fucking drives me insane.
>>109719602A few days ago Tibo was talking about using the internet mostly through Codex now. They are going to push it more and more into being all-in-one app. Just like Elon first said he wants to do some years ago. Really all of them want to take over everything.
>>109719657>number go upwoah
>>109719648>1AM PTany news of a reset so I should switch to FAST
>>109719629you should consider trying fable if astra isn't the new SotA when it releases. fable 5.1 is really good at compressing codebases and unspaghettifying things, and you mentioned those were big problems for you. don't pay per token, the rates are insane, just upgrade your sub for one month and try it.you should still wait for astra and see how that benchmarks in comparison though, it may end up being cheaper and better. give it time.
I'm astral projecting
>>109719602Looks like they want to put some serious amount of computation on your client. To do what?
>>109719648>4 am ET on a fridayalright see you next week
>>109719692what do you see?
>>109719602Astra seems to completely rework how reasoning works.One obvious next step for improving speed and latency is to basically not have a harness work the model but codex be more of a fast tool call tunnel, where the actual harness runs on the provider's server and the local app is just backdoor for tool calls, that require local access.Like there is no need for a search call to make the roundtrip.
>>109719319wait what, why??
>>109719695more subagents
>>109719732just use Pi
>>109719680> and you mentioned those were big problems for youMore like I hope it's not a problem, because I can't in all honesty, judge it at this point.It werks, so far, and that's that.>just upgrade your sub for one month and try it.I was considering this, but at this point, I just can't justify the cost. It's a lot and I didn't even make a dime yet.I hope I can get a vertical slice soon and open the patreon et all stuff so at least I know there's some interest. I don't expect it to pay for itself (quickly), but I need to have a sliver of hope before throwing the big bucks at it.>fable 5.1 is really good at compressing codebases and unspaghettifying thingsAs a note for the future, how should I go through it?I mean, I assume it's not as easy as "go through the codebase, unspaghetti things". I was thinking I would make it prepare the high level docs from current stuff, then docs and dictionaries for each functionality and then go to some map of classes, functions etc. coupled with those.Sounds reasonable?
>>109719699Very bad take. Web search is not going to help you on a project other than the occasional Arxiv paper or GitHub repo. It's not remotely a bottleneck and too much search drowns the model in noise.>>109719695It's just marketing. That guy is technically incompetent and doesn't know what he's talking about. Weeks ago he twitted a method to extend codex's context that hadn't worked since GPT 5.4. He's just very good at manipulating people aka marketing.
>>109719773>I mean, I assume it's not as easy as "go through the codebase, unspaghetti things".but it is. tho expect to burn lots of tokens.https://x.com/bcherny/status/2027534986178662573
>>109719657you made me check
>>109719807jfl at this imbecile
i love llms
>>109719806Codex bros we need this
>>109719657"this changes everything" lmao
>>109719773i'm not kidding, it literally is as easy as "unspaghetty my code" for fable 5.1. you should still make a backup of what you have right now, never ever ask for big sweeping changes without backing things up in case something goes wrong. but ask it to fix it for you and describe all your concerns you told me to claude.
>>109719848>you should still make a backup of what you have right nowyou mean just have the code in a git repo, right?
>>109719848>backing things upit's called git>You use Git — right, anon?
>>109719853>>109719855git isn't really necessary when you can just tell fable to undo what it did 99% of the time
>>109719848cc has checkpointshttps://code.claude.com/docs/en/checkpointing
>>109719853>>109719855hivegit
>>109719853>>109719855You're joking, right? You know these things can delete your whole home folder
>>109719862>fable, ctrl+z make no mistakes
>1h downtime>no reset>no astraI'm gonna sleep
i cant use the word beautiful anymore, fucking jeets
did tibo change his pfp after some of you jerks were bullying him here
>>109719869>make no mistakesverification rituals are anti-patterns for fablehttps://x.com/RLanceMartin/status/2094854835854295296
>>109719853>>109719855hey reddit i don't know if you've read any of that reply chain before you tried snarking like niggers but the guy being helped literally doesn't know anything about code at all. "git" is probably not even a word he knows. "back it up" is the easiest way to describe it to a nocoder, who will ask claude and then can be handheld through it.
>>109719923>make 2 mistakes
Good show, faggots.
>>109719923do you really expect me to read all that bullshit honky boy
>>109719923Glad we are back to prompt engineering.
>>109719878>Indian outing himself Not tricking anybody with that LotR pic.
>>109719806"skills" and all it's derivatives is probably the worst of all the snake oils sold to vibecoders
>>109719868Yeah so you push your commits to private GitHub repos lmao. The elites don't want you to know this but you can store terabytes on GitHub for free as long as you keep it in small enough files.
>>109719973>jeet jeet rajeesh indian brownchudai
>>109719926>hey reddit
noc*ders disgust me, buncha larping retards, please leave, all you make is poo
>>109719926I'm the newfriend, but I'll be happy to inform you, I'm not completely incompetent and have git and a repo set up with a side branch specifically for ai development; and AI has a strict "do not touch" policy regarding commits.But thank you nonetheless! I really do appreciate all insight!
>>109720018Stop shitposting.
>>109719280>>109719380>no gpt 5.6 modelsthe fuck shitty ancient garbage is thiswhy are you shilling this?
will astra finally save us from anthropic?
>>109719978I'd rather not upload all my files unencrypted to Github but that's just me
https://openai.com/index/playco-game-prototyping-with-astra/
>>109719978>>109720040Can you partition your encrypted backups into small chunks and commit them, or will they ban you?
Astra mogs fable
I told it to design a cipher and Fable punted me to Opus 5, which punted me to Opus 4.8, which respectfully declined without an explicit safeguard hit.
>agi is hereAnd its still fucking slop
>>109720058Yes – rot13 is a good choice.
>>109720040Yeah have to really protect that stuff AI can generate in minutes. Very sacred.
>>109720046vibin
>>109720064I use ROT26, it's twice as good as ROT13
Will Astra finally have 1m usable context at a normal price?
>>109720032truth hurts
>check local inference providers>Kimi K2.6>Qwen3.5 397B A17B FP8Yeah... Compliance is hard.
>>109720085I think context space is more an infrastructure cost and memory space constraint than model power, they might boost it anyway though
tell me it isn't true
>>109720098boring
>>109720098where the FUCK is my astra
>>109720098there's no way this is true. Fable/Gemini(lol) launched Day 1, and OpenAI has a shit load of compute.
>>109720051I don't think they care what people upload to GitHub, you can upload malware, you can even reupload stuff that was taken down from GitHub in a DMCA takedown. Seems like you should limit yourself to 1 GB per repo but can have 100,000 repos lmao.https://github.com/SameerMatoria/Git_Cloud
>>109720121openai has been sucking hard latelycopex poster won
Polymarket still 93% for this day.
>>109720125Fucking lol.
I just did KYC with ChatGPT for Daybreak access. I can't fucking afford to miss astra for 3 days
it's fucking realhttps://www.cnet.com/tech/services-and-software/openai-gpt-6-astra-release-ai-agi-chatgpt/https://venturebeat.com/technology/welcome-to-the-agi-era-openai-launches-gpt-6-astra>Astra begins rolling out Thursday to enterprise customers with OpenAI's gated access program, Daybreak. OpenAI says it will become available over the coming days to ChatGPT Plus, Pro, Business and Enterprise customers, as well as through the OpenAI API and cloud platforms including AWS Bedrock and Microsoft Azure.
>>109720098https://www.ft.com/content/55ab40c0-59e2-4c0b-97c9-4f4f5a71a8bb?syn-25a6b1a6>The model will initially be rolled out to a small group of businesses to allow time for them to address cyber security concerns before becoming widely available “over the coming days”.>Astra will cost as much to use Anthropic’s leading model, the take-up of which has plateaued since it was launched as users turn to cheaper alternatives> Greg Brockman, OpenAI’s president, said the new model “represents a generational leap in capability” and that it could be defined as artificial general intelligence — roughly defined as a point at which AI tools surpass human capabilities across a range of cognitive tasks. “Everyone has a different definition of AGI...it’s a grey, fuzzy thing. But I think when we look back people will think it’s about this time and about this model,” Brockman said.
>Astra scores 97.6% vs Fable 5.1 88% on FrontierMath Tier 4 v2>Astra scores 74.1% vs Fable 5.1 67.4% on DeepSWE v1.1.>Astra scores 95.9% vs Fable 5.1 84.3% on BenchCAD.>Astra scores 96% vs Fable 5.1 94% on GPQA Diamond.Astra scores 98.6% vs Fable 5.1 97.5% on ARC‑AGI‑3
>>109720046>“If you have 10 ideas for a game, you can do all 10 and actually play them and see how they would feel rather than just imagine.”Reminds me of the guys who made world of goo
>>109719145get codex $20 and/or opencode go
artificial analysis waiting room
>>109720175IT'S FUCKING OVERTHE NUMBERS WENT UPTHEY WENT UPPER THAN ANTHROPIC
>>10972016774 on DeepSWE would be disappointing, that's just Sol.Benchmarks don't mean that much, but I wonder if the model is mainly for cyber security, I just want it for code.
>>109720167This is gonna becoke the standard. Premium corpos get a few day do check for security issues.
Muse spark 1.3 beats Astra on deepswe 1.1
>>109720175OpenAI takes the lead with a marginally better score in some benchmarks.
https://openai.com/index/gpt-6-astra/
>98.6% on ARC-AGI-3
Now I'm sad, why did they blue ball me like that.
>>109720204deepswe is benchmaxxed these days, better wait for a new benchmark that discriminate models properly
>>109720230I will believe it when I see it for myself
Astra Pricing$10 million input tokens$50 million output tokens
>>109718201i hope they recognize a significant part of their market is people frustrated with safety claude-tism
can they fucking at least release gpt image 2.5
>>109720253>going from 8% to 98%fake
>>109720253>BenchCAD 95.9%holy shit. PLEASE FUCKING RELEASE I NEED TO PRINT AHHHH
>>109720098im cancelling gpt pro sub regardless of how gpt-6 assturd is
>>109720230>marginally better score>10 point jumpspick one
>>109720253The DeepSWE numbers are a bit random, at higher reasoning for instance Opus scores much higher.
So it looks like I will still be stuck in abusive relationship with claude.
>>109720098how is the guy who whined about anthropic's safetygroiding coping now that openai is doing the exact same shit LOL
>>109720268>CADdoes it make STLs or what's that about
I think they embargoed the news, and then they failed to put up their own blog post synchronized to the scheduled embargo releases, probably because of the outages they're having today.Reuters announced at 2.03 p.m.All the news articles say that OpenAI announced it in a blog post, of course.
>>109720361I'm not sure what the specifics of that benchmark are, but I've had good results with Sol + Fusion MCP.
just knocked my drink over on my desk, covering myself, which model best prevents this
>>109720392glm 5.3 flash
full benchmarks are out. it scores 67 on the artificial analysis coding agent index. note that's different from the general intelligence index
>>109720405the coding index, for comparisonhttps://artificialanalysis.ai/agents/coding-agents
>>109720405Does it say what effort level it uses? Because all the numbers for the other models make no sense.
I'm gonna ask Astra to make the code that Sol shat out more simpler and readable lmao
>>109720405Fable 5.0 mogs Fable 5.1 on some tests?
>>109720432nope. openai actually removed it very shortly after, so we'll see what's uphttps://openai.com/index/gpt-6-astra/
>>109720405>>109720427>AGI is here>and it's not as good as fableif the numbers weren't leaps and bounds over fable they should not have said in interviews that it's "basically AGI already" that's absurdi'm hoping when i test it it'll be better than fable but this is incredibly disappointing
>>109720372>autodesk wants my moneya cold day in hell
man, this was kind of embarrassing, huh? will tibo give resets as apology?
Diminishing returns, huh? AGI cancelled.
>running multiple claude sessions 18 hours a day since the reset>one follows the plan>one doing cleanup items>one planning the next phase(s)I always keep them on a tight leash and micro managedon't know what this is doing to my mental in the long run, so maybe it's a good thing I'm losing 17% weekly soon
>>109720456Some of the numbers are very good, it seems to be very strong in non coding tasks.
What causes models to think and not respond. Sometimes I just end up in a situation where it won't respond for a few replies over the span of minutes. No output just "Thought for 2 minutes" status messages. The fuck is he thinking about.
>>109720501What model at what reasoning level? Some models love just thinking about stuff.
>>109720500I only use AI for coding.
>>109720175I wasn't expecting them to nuke them from orbit like this
>>109720456It beats Fables 61? Thats coding index, not agentic
Overhyped
Screenshot of the OpenAi' announcement of Astra:https://ibb.QQQ/k2fB5wScQQQ=co
>>109720510I've had this problem with a lot of them for a while now. Usually with long/persistent sessions. Just sometimes it gets silent for a while.I vaguely remember Gemini doing this to me sometimes, but usage on that was limited.I've encounter this more often with the Chinese models, but that's because I have significantly more usage of them due to their costs.Every time it happens I feel schizophrenic.Man types at computer terminal, genuinely expecting it to respond.
>>109720511Same, but very strong math capabilities could be interesting. Maybe it could come up with completely new algorithms, which would then enable writing more complex software on top of that.
>>109720175>>109720519I hate Anthropic more than I hate OpenAI but the difference is probably Anthropic serves the same model they run the benches on. OpenAI always does this same bullshit of spending a million dollars per benchmark and the model you are actually served is run with 0.000001% the amount of inference compute than the models they bench on.
>>109720534virus
>>109720544oh good catch, nevermind. i'm excited to try it again.
>got fable working efficiently and using opus slaves for labeling>one week to fix the code before they nerf 5.1>can always have codex refactor the code if it's too claudeified later since the tests will ensure it works the sameit's cooking time
>>109720534>>109720552fucking catbox is broken againhttps://litter.catbox.moe/ep2626vkl71vf81u.png
>>109720552>>109720566tangentwhat the hell is going on with catbox mayneI hope the admin is okay and not getting fucked up somehow. this shit always broken now. I'm glad he hosts it for free though, even with people like me complaining about it
>>109720570more people are using it and shit (hdds, ram) costs way more now
Astra better come soon because Sol is completely retarded today
>>109717519Wow I remember when this general could barely be kept alive and now it hits 500 replies in 6 hours?
>>109720570demand issue as people share bigger and bigger files (videos and images)
>>109720602discussion around fable and astra accelerated replies per minute. atm it's one of the fastest generals on the site
>new version>pet gets a keyboard shortcut>still can't move the pet or click the buttons on it
>>109720566They're calling it the worst screenshot of all time
https://astratest.codergautam.workers.dev/GPT-6%20Astra_%20A%20new%20generation%20of%20intelligence%20_%20OpenAImirror
>>109720602masha is so fucking cute
>>109720456If it's anything like Fable vs 5.6 Sol, it'll be if you're groping about for the AI to figure out what your half-baked prompt actually means, correctly guess at your requirements and specification along with making a really nice web UI then you'll prefer Fable. Otherwise Sol is already preferable.
NO ASTRA TODAY
Reminder that bel will make astra look like a toy in december
where is gpt image 2.5?
>>109720635this slow rollout bs makes it look like they are just now seeing whatever anthropic saw at the start of this year
>>109720602most traffic is modelfagging and corpo war shitpostingtalking about projects and coding has all but disappearedseems to happen to all generals in the last two years. Bots and zoomers have completely taken over
>>109720548I can pose biological and chemical shit to gpt without it thinking im a war criminal which im not claude insta calls hr on the mention cyanide
>>109720647>Otherwise Sol is already preferable.the schizo woke up again
>>109720661buy an ad Sam
>>109720625werks on my machine
astra better fix every problem in my project within minutes
>>109720625touristsissy...
>>109720566thanks bro
https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf
>>109720736oh, it'll fix it in minutes. no less than 180
>>109720742>Dynamic Mental Health Benchmarks with Adversarial User Simulationsholy shit just in minecraft anyone that will need this
>>109720714Sol's reasoning is moderately better than Fable's. Fable's general ability to adapt to PEBKAC is much better than Sol's.
So the Astra launch today was a nothingburger?Never again will I listen to xitter accounts with anime profile pictures
>>109720755>it's realat what point do you just let natural selection kick in LOL holy shit>>109720765yeah sure man whateverask tibo for another reset
>>109720774It's a limited launch, so nothing until maybe 1-3 weeks for the rest of the users.
thread uphttps://x.com/OpenAI/status/2095595741528125780blogpost (currently 500 for me)https://openai.com/index/gpt-6-astra/
>ARC-AGI-3 tests how well agents learn as they solve unfamiliar interactive tasks. GPT‑6 Astra saturates the eval, scoring 99.9%. The average human tester scored 48%. GPT‑6 Astra was measured with our responses API harness, which better reflects real-world performance than the original benchmark harness, which discards past reasoning and past messages. With this harness, we estimate Sol would score in the ballpark of ~30%.HOLY COPE
>>109720817Used to be 404 for me a while ago today.
>>109720833>we estimate Sol would score in the ballpark of ~30%.Why didn't they just run the test with sol.
kek benchmarked copex slopnot even gonna resub to try ut
I hope chinks already started the distillation somehow. I want it for half the price on multiple providers.
>>109720833>we estimate Sol would score in the ballpark of ~30%????
I need it now
>>109720755These literally are a negative. I want them to be as psycho as possible
>>109720881Reality is it's only really better at making music and navigating mundane office tasks and they're making everything else up
NEW TIBO>We are starting to release GPT-6 Astra and we are doing it as carefully and quickly as possible. It was very important to us that we bring it to all Plus users and not only Pro, Business and Enterprise. >It will take a few days for the rollout to complete and behind the scenes many novel systems will operate at scale for the first time and we are bringing a lot of compute up.>It is pure magic.I'm thinking some normal people might see it as soon as tonight
>>109720901I'm also somewhat optimistic, although not quite as optimistic as you. I think it will probably be some time this week.
>>109720848Because it's a shitty fucking benchmark, that's why. Imagine making a human take an exam, but part of the instructions was that they're only allowed to use the left side of their brain or they get disqualified.
This is basically the end, right? Now that AI of such power is widely available, employing humans for anything other than physical labour is just retarded. You will immediately lose if you hire meatbags for any office job, while your competitors are vibing with Astra.
>>109720915there's only tomorrow and a friday launch is already weirdI doubt they will roll it out on a weekendSo I'm guessing plebs will get it next week at the earliest
>>109720928Why would you hire meatbags for office jobs since the entirety of 2026 to begin with I dont understand
>>109720928Yes, and prompt manglers like ITT will lose their job too.
lol
>>109720929Hmm, maybe you're right. I was thinking that maybe it will roll out over the weekend, since it's already officially launched and the last few steps would be easy, but maybe not.
>>109720817>all those seething jeetsWHERE IS ACCESS TO ASTRA SAARMOTHER BLAADY WORST RELEASE EVARREDEEM THE ASTRA BLAADY BHENCHOD
>>109720953fuckyou made me check faggot
>>109720953KEK LET THE COPE(X) BEGIN
>>109720928holy marketerwe've had astra for months (fable) and this hasn't happened. wait another model generation before you start wheeling out the "WORK IS OVER" thing.
>>109720901People are seething in that thread. Would have been nice to have it today, but OpenAI never promised that and I didn't get my sub for Astra specifically, so what's the big deal?
>>109720973Inertia to making people keep jobs and old standard shit running on human hands and people afraid of the jobocaust are keeping things slightly together for a while but reality is we will eventually need to restructure what jobs are actually worth having humans for
>>109720971The bottom part is better in Sol's version.
>>109720942Because at the start of 2026 models were still too retarded to replace absolutely everyone. SWE yes, I would say even in 2025 it was over for them. But to actually replace the entire company, this is finally the moment where it's not only possible, but also financially viable. Astra is the only employee you will ever need.
>>10972092890% of companies already exist at a level where they don't actually do any meaningful work and are more busy with managing themselves than creating any product or valueyou could've fired 50% of the entire white collar workforce 10 years ago and literally nothing would happen to the productivity of those companies
@thsottiaux Why did OpenAI lie and hype up today as the release day when it's a rolled out release? Everyone thought we'd all have it available in Codex today. Do you realize how pissed everyone on Twitter and Reddit is?
>>109720971they both pretty ass
>>109720971A pro 3d artist would make you this in under 40h but exactly as source
>>109720996I disagree, something like Opus 4.5 still wasn't good enough to do all software engineering, maybe it was for easy web apps.
Astra is, good?
>>109721007So how long did AI take? Couple of minutes?
>>109721012What's a continuous conversation harness?
>>109721016Did it make it exactly as source? Different benchmark here
>>109721027the re-education harness from clockwork orange
>>109721027Must be something like the brainwashing thing in A Clockwork Orange.
>>109720998No we need hr nancy and stacy
>>109721034That is a good question. Maybe the AI 3D models look like nonsense when you rotate them.We will never know.
>>109721059I'm a pro 3d artist I work with other 3d artist I'm not making that up. 3d ai is still kinda not quite there yet. It's not that easy
>GPT-6 Astra scores equal to GPT-5.6 Sol in the Index at 61. This is 5 points lower than Claude Fable 5.1 (max with fallback). The model also trails Meta’s newly released Muse Spark 1.3 (max).
>>109721016Meta gpu setup point clouds to consumer gpu point cloud work from meta is at a 100x speed disparity I've run the tests myself
https://x.com/sama/status/2095601211869421726>We are working towards getting Astra in everyone's hands as quickly as we can; I know it is frustrating and I appreciate the patience. It should be generally available within the month.
>>109721094>month:(
uh...
>>109721094The good news is, we've apparently had AGI for months: Fable!
>>109721083>>109721108>ITS REALOH NO NO NO
>>109721108>on par with opus 5unreal. should've delayed it man. gpt image 2.5 will be cool though at least
If someone were to post a custom solution to navier stokes tomorrow how much would these models progress spikes look like
You shouldn't care about the ArtificialAnalysis benchmark, it's basically 100% coding. GPT-6 is obviously a way better model IN GENERAL which is what matters
>>109721109But seriously Fable could be called AGI. Fable is generally very good, but having a slightly cheaper version of the same would've been nice. Even if it's the same per m token, the fact that it's supposed to be 100% of your account makes it already almost half as expensive.
>>109721124given openai is just now acting like anthropic 9 months ago, i have sincere doubts
>>109721124Coding is all that matters. General knowledge is something any model is capable of
>>109721140Its all marketing nigga they made it better at doing office boy shit and thats it
>>109721108AA lost all its meaning months ago. Trash benchmaxxed chink models keep scoring high. Muse Spark higher than Fable? lmao
>>109721108Ah, this is kinda sad. It's nice that it's good at math and so on, but coding still is the main thing for me. I was already slightly disappointed with Sol but I thought Astra would be much better.
>>109721150using the marketing tactic your only real competition already used seems daft
>>109721124i don't know if they give you different scripts depending on where you're being paid to post but you're posting this in the vibe coding general. i think you need to pick a different cope.
>>109721160No it isn't all of the ai owners are trying go pose ai as superterror and superpower to bring people in
>>109721094>from within days to within the month
>>109721173Artificial analysis scores dropped and now they need to scramble to get gpt 6.1 ready to release so it can be on par with fable 5.1
if open ai spent their marketing budget on actually making a better product instead of trying to gaslight me into buying their worse one i would actually spend money on their 200x plan
>>109721189>200x planoil sheik anon...
>>109721124You are a fucking retard. Intelligence Index flatlined and only 2 out of 11 benchmarks in that test code (scicode and terminal-bench). The rest test is GDPval-aa v2, a white-collar benchmark and etc
Oh boy more benchmaxxxing wake me up when I can actually test it in the real world
>>109721226First the big boys have to make sure you won't hack the world with it
New thread:>>109721250>>109721250>>109721250
