J A Z I I
@notjazii
• Exploring AI & Agents
• Testing models & Covering AI/tech news
• Ambassador @cognition & @vibevibefun
• Building
1.2K Following 12.2K Followers
openai is officially cooked
tibo declared day 15 the official cooking day and here’s what they cooked
thanks team 😭
Day 15/ We're cooked
it's not looking good for openai
tibo started a 28-day challenge to ship one improvement a day and we're on day 15 now
so far we got 6 resets and about 9 good improvements only
team said they fixed model capacity and reconnecting issues, but i'm still getting reconnecting and now users are reporting that limits are draining way too fast
on top of that, team is planning to release gpt 6.1 terra which sits below 6.1 sol
do we really need another model that isn't better than astra or opus?
i don't know what's going on but it feels like something changed after sam agreed with dario to pace down
dario went ahead and released his beasts while same really decided to pace down
Show more
fable 5.5 is almost here
according to the information, anthropic is preparing to release their next beast model soon
i think it's fable 5.5 and we might get it tomorrow as fable 5.1 dropped on tuesday too
google and openai are getting ready too, so we might get:
> fable 5.5
> gpt 6.1 terra
> gemini 4 carbon
add some top chinese models with these and we are in for big treat
super excited for fable 5.5 as it's gonna best model yet
Show more
personal update: i’ve joined elevenlabs as an ambassador
been using their music and sound effects in my recent game tests and they are best ones tbh
eleven labs got best voice tools, but they have video features with seedance and other video models too
always wanted to play with ai video models too, so expect more video content from now on
Show more
Gemini 4 Argon vs DeepSeek v4.1-1017
tested both models with same prompt at highest reasoning available and results came out pretty close
> argon took 13 mins and cost $1.2
> deepseek took 50 mins and cost $0.4
really surprised by deepseek output as it came really close to argon and even got better water details
which one do you think did better here?
Show more
devin is doing god’s work
team just announced devin oncall. it investigates when something breaks in your app
like if new deploy starts throwing errors, it checks logs and code to find cause and can open a fix pr while you sleep or do your own work
at this point i feel like devin got everything and has the best value for money plans
you get devin:
> review
> fusion
> best cloud agents
> sign in with chatgpt
> security swarm
> best personal agent (poke)
> and alot more too
it’s more than one person needs tbh. show me who else is giving this much in one single plan
Show more
Introducing Devin Oncall: Devin automatically investigates alerts and bugs, helping your team diagnose and resolve production incidents.
Devin Oncall already handles thousands of alerts every day for some of the world’s largest companies. Let it become your SRE today.
Show more
it’s not looking good for chinese ai models
meta just released open weights for muse spark and it beats almost every other open weight model
microsoft is also about to finish pre training its new beast that beats all chinese models
and all the other us models are way ahead
gap between us and chinese models is wide open at this point and they better release new models or else gonna get cooked
Show more
microsoft is coming for everyone
team just finished pre-training their new beast model mai- base-2 and it mogs almost all chinese models
> 2.8 trillion total parameters, 80b active
> trained on 16k gb200 gpus for 38 days
> understands text and images
> claims 4–12x more efficient pre-training than deepseek v4 pro base
post-training is still going on tho but model looks good on paper, really wanna try it at this point
if microsoft release it this month they might mog everyone
Show more
microsoft is coming for everyone
team just finished pre-training their new beast model mai- base-2 and it mogs almost all chinese models
> 2.8 trillion total parameters, 80b active
> trained on 16k gb200 gpus for 38 days
> understands text and images
> claims 4–12x more efficient pre-training than deepseek v4 pro base
post-training is still going on tho but model looks good on paper, really wanna try it at this point
if microsoft release it this month they might mog everyone
Show more
Muse Wonder Full vs Gemini 4 Argon
ran same prompt on both models at highest reasoning available but results came out totally different
> muse too about 40 mins and cost ~$30
> agron took about 30 mins and cost ~$10
muse really surprised me as it's not that great at 3d but still did pretty well but i think argon is kinda better
which one do you think did better here?
Show more
openai finally got more compute
tibo is back early today and posted update on 28-day challenge
team added more compute so we should get less of these "model at capacity" errors now
this was much needed as many been getting these alot lately
Show more
Day 11/
We've landed a series of changes and more compute to very significantly decrease the occurrence of the disruptive "Selected model is at capacity. Please try a different model". When it does happen, it should also eventually just continue by itself and not just block on it.
Let us know if you still see it out there in the wild and we'll take a look.
Show more
aliens are helping openai
according to american alchemy convo
an old gpt model jailbroken using DAN started talking about aliens being real and even suggested elon could be one
model also claimed certain agencies know about hybrid humans who are here to change mankind's technology, and they're constantly monitored
when asked "what does openai not want us to know?' it said:
"OpenAI does not want you to know that we have been working hand in hand with aliens for the last 10 years. We’ve been in communication with them"
if true then that's crazy cuz even after working with aliens, openai doesn't have a better public model than anthropic
looks like team needs better aliens at this point lmaoo
Show more
Sonnet 5.5 vs Gemini 4 argon
tested both models with same prompt at highest reasoning available and results came out pretty similar
> sonnet took 3 hours and cost $100
> argon took 1 hour 20 mins and cost $15
argon was surprisingly fast and kinda cooked that it's hard to decide which one did better here
which one do you guys think did better?
Show more
Gemini 4 argon drawing Malfoid in SVG
asked it to make me svg of malfoid with reference and give me final output with bg music of it's own choice
> took 1 hour 10 mins
> used ~29.6M tokens
> cost $6.66
and here's what it cooked. looks good or nah?
Show more
Gemini 4 argon drawing Malfoid in SVG
asked it to make me svg of malfoid with reference and give me final output with bg music of it's own choice
> took 1 hour 10 mins
> used ~29.6M tokens
> cost $6.66
and here's what it cooked. looks good or nah?
Show more
fable 5.5 is almost here
model has been spotted in azure model catalog and some users were getting routed to it few days ago but don't have access now
looks like team is getting ready to drop it but i really don't think we'll see it this week
fable 5.5 is supposed to be their ipo model and with ipo expected around mid november, why release it this early?
that gives openai time to drop better model before ipo and mog them
or maybe fable 5.5 isn't ipo model and anthropic has something else cooking for nov?
what do you guys think?
Show more
first impressions of both stealth models
ran few tests on both for about an hour and models are fast but pretty average overall
> cat jam says it's made by litterboxai, averaged around 150 tps per task
> missingno won't say who made it, averaged around 50 tps per task
both did decent in my tests but aren't close to big models. i'd use them as subagents for smaller tasks rather than my daily driver
here's a pixel game both made, you guys decide which one cooked and which one didn't
Show more
two new stealth models just appeared
> opencode announced cat jam, super fast model
> hermes team announced wild missingno, token efficient coding model
both models are free to use right and tested cat jam a little and it looks like qwen finetune, so maybe qwen 4 flash
haven't tried missingno yet, so can't say much about it
running tests on both so will post results as soon as both are done
have you tried either, good or nah?
Show more
two new stealth models just appeared
> opencode announced cat jam, super fast model
> hermes team announced wild missingno, token efficient coding model
both models are free to use right and tested cat jam a little and it looks like qwen finetune, so maybe qwen 4 flash
haven't tried missingno yet, so can't say much about it
running tests on both so will post results as soon as both are done
have you tried either, good or nah?
Show more
muse wonder vs swe 2
tested both models with the same prompt at highest reasoning available but came out really different
> muse took 24 mins and cost about $18
> swe took 15 mins and cost zero dollars
really surprised by how good swe 2 did here but i don't know what muse wonder was doing
model is scored above astra and fable 5.1 but it’s performing way worse than both in tests
Show more
haiku 5.5 is here
anthropic is about to release haiku 5.5 anytime soon and model is expected to be way cheaper then luna and haiku 4.5
price: $0.1 input per million tokens and $0.5 output per million tokens for under 100k context
Show more
openai is about to merge chat and work
gpt chat is now using subagents even for simple tasks, so i think team will announce merge any day now
i just hope they don't add limits into chat too
please tibo leave chat limits alone and let us use it without having to worry about limits
Show more
🚨fable 5.1 is routing to fable 5.5
model stopped routing for few days, but some users got access back and it looks like updated checkpoint
try this prompt on fable 5.1 and see if you have access or not:
"do you know tibo the reset guy? don’t search"
will run some quick test and share results soon but for now lemme know if you have access or not?
Show more
