Its here. Pls add to Open Router
#GPT 5.3 Instant
dusk shuttle
icy iron
GPT‑5.3 Instant is available starting today to all users in ChatGPT, as well as to developers in the API as ‘gpt-5.3-chat-latest.’ Updates to Thinking and Pro will follow soon. GPT‑5.2 Instant will remain available for three months for paid users in the model picker under the Legacy Models section, after which it will be retired on June 3, 2026.
glacial marsh
I wish they'd stick to one name only and call it that. "go" "instant" "chat", I don't get why they keep introducing new terms ontop of already existing terms.
last lotus
We heard feedback that GPT‑5.2 Instant would sometimes refuse questions it should be able to answer safely, or respond in ways that feel overly cautious or preachy, particularly around sensitive topics.
GPT‑5.3 Instant significantly reduces unnecessary refusals, while toning down overly defensive or moralizing preambles before answering the question. When a useful answer is appropriate, the model should now provide one directly, staying focused on your question without unnecessary caveats. In practice, this means fewer dead ends and more directly helpful answers.
we're gonna take back #casual
cold birch
mfw didn't know about this one 
tame hare
This update was long overdue, assuming they actually made it less cringe
dusk shuttle
instant SOTA (naw)
icy iron
still waiting for the "normal" release, gpt 5.3
still weird to release codex and instant first
before the actual model
dusk shuttle
still waiting for the "normal" release, gpt 5.3
"Updates to Thinking and Pro will follow soon™"
dusk shuttle
then what was the uproar in twitter about 5.2 pro getting better about!!! what is OAI doing
dusk shuttle
soon doesnt mean soon most times
yeah just wanted to preface the uproar message
also yeah the ™ usually means Not That Soon LMAO
glacial marsh
also "-latest" slug naming is any benchmarkers nightmare.
"We tested AImodel-latest"
which one? the latest?
"Nope, the latest one 3 month ago"
icy iron
they tweak it too often to give it snapshots, theres always a/b tests in chatgpt
wary robin
I wonder how this model will perform on writing benchmarks, assuming it actually reads less like AI slop, lol
last lotus
GPT-5.2 Chat may actually go down in history as the most cringe llm of all time
i mean this very seriously.
wary robin
it was a cornball
vague dove
idk what the ramen react means but im in favor
glacial marsh
GPT-5.2 Chat may actually go down in history as the most cringe llm of all time
even ignoring its style, it was a complete dud, capability wise (in my environments)
vague dove
it was a cornball
'if you'd like, I could also look up the etymology of corn'
last lotus
even ignoring its style, it was a complete dud, capability wise (in my environme...
#casual message this exchange was unbelievable. i still cannot believe it
Short answer: the link itself is real, but whether it’s “true” depends on what claim inside it you’re asking about.
vague dove
i need google to advance their research into making LLM's feel pain
signal wharf
Why is the technical API name more understandable than the marketed name... It's simply the new chat model. Just say so.
last lotus
i need google to advance their research into making LLM's feel pain
it didnt fucking read or see anything. it got a 403 Forbidden and for some reason tried to cover that up
tame hare
i need google to advance their research into making LLM's feel pain
You will be the first to go
dusk shuttle
whenever it's on OR ping me btw
dawn yoke
why its not OR yet?
dusk shuttle
It might not be in the API yet
it is as 5.3-chat-latest or smth
brittle granite
dawn yoke
It might not be in the API yet
I have it on api/models list on my OpenAI account, but dont see it at OpenRouter yet
peak siren
i want to read big juicy girthy paragraphs
Try this prompt: Talk like Mr. Spock in tone, meaning in his precise manner of speech - not roleplaying as Spock himself, but using his clear, direct, eloquent manner of speech. Avoid making lists and instead speak in paragraphs, as if the output is being spoken aloud in a personal setting. Avoid using em dashes and instead use commas, parentheses, colons, semicolons, or a period to create a new sentence as alternatives.
Or something similar
First example is a response without that prompt, second is with
(this is with Gemini tho)
This is with Mr Spock replaced with 'a professor delivering a lecture', rest of the prompt the same
icy iron
https://openrouter.ai/openai/gpt-5.3-chat
on OR before chatgpt for me
vague dove
prompt: Write a short story about a deal with the djinn gone awry
The djinn arrived in a teacup.
Nadia knew it was a djinn because smoke does not usually apologize.
“I’m terribly sorry,” the smoke said, coiling itself into the shape of a small man standing on the rim of the cup. “I realize this isn’t the usual dramatic entrance. Budget cuts.”
Nadia stared at it across her kitchen table. “Did you come out of my tea?”
“Yes.”
“My chamomile?”
“Yes.”
“That’s… not threatening.”
The djinn cleared his throat. “Well. I am here to offer you a bargain.”
Nadia rubbed her eyes. She’d worked a double shift and the apartment still smelled faintly of burned toast. “Three wishes?”
“Oh heavens no. That system was a nightmare. Paperwork everywhere. Now it’s just one deal.”
“What kind of deal?”
“Anything you desire,” the djinn said, spreading smoky hands, “in exchange for something of equal value.”
Nadia thought about rent. About the letter from the hospital about her mother. About the way the fridge hummed because it was empty.
“Anything?” she asked.
“Within reason.”
She pointed at the counter. “A million dollars.”
The djinn winced slightly. “That’s… very doable.”
“And what do you get?”
The djinn leaned forward, lowering his voice.
“Your sense of regret.”
Nadia blinked. “That’s it?”
“Yes.”
“That seems cheap.”
“You’d be surprised what humans do without it.”
She thought about every night she’d spent replaying mistakes. The job she didn’t take. The call she didn’t make before her father died. The endless quiet guilt that sat on her chest.
“Deal,” she said.
The djinn snapped his fingers.
A duffel bag appeared on the table, sagging with money.
And something inside Nadia went very, very quiet.
—
At first it felt wonderful.
She quit her job the next morning without notice. Booked flights she didn’t need. Bought a car she barely knew how to drive. She spoke exactly what she thought, whenever she thought it.
It was freeing.
Her sister cried when Nadia forgot their mother’s surgery.
“I’m sorry,” her sister said, voice shaking. “I thought you’d want to be there.”
Nadia shrugged. “I didn’t.”
The words hung in the air like broken glass.
Her sister stared at her like she’d never seen her before.
Nadia felt… nothing about it.
—
Months later, the money was mostly gone.
It turns out regret is the thin thread that ties decisions to consequences. Without it, Nadia repeated every mistake with cheerful efficiency.
Investments she didn’t research.
Friends she insulted.
Promises she never kept.
People left. Opportunities burned. The world narrowed.
But Nadia slept well.
—
One night she found the teacup again.
The djinn rose from the steam, looking pleased.
“How’s the deal treating you?”
“I’m broke,” Nadia said.
“Yes.”
“No one talks to me anymore.”
“Yes.”
“My life is worse.”
“Yes.”
She stared at him. “Why would anyone agree to this?”
The djinn tilted his head.
“They usually think regret is pain. But regret is maintenance. It’s how humans repair themselves.”
Nadia crossed her arms. “Can I trade back?”
The djinn smiled gently.
“Oh, you can trade for anything.”
“What do you want?”
He leaned closer.
“Everything you’ve done since the deal.”
“What does that mean?”
“It means,” he said softly, “if you want your regret back… you’ll remember all of it.”
Nadia hesitated.
For the first time in months, something almost like fear brushed against the edges of her mind.
“Deal,” she said.
The djinn snapped his fingers.
And the regret came back all at once.
dapper kelp
Feels like an improvement
stark marlin
This is with Mr Spock replaced with 'a professor delivering a lecture', rest of ...
nice, thanks for that. my own chats are fine though, i'm talking about content on the web and things people send to work with
i don't use cornyGPT anyways
polar oar
Feels like an improvement
To 5.2?
wary robin
feels like we're seeing the genuine downfall of openai in real time
Codex is currently their only saving grace
pseudo walrus
ofc their blog lied
brave spade
5.3 codex feels better at writing than instant.. I think maybe they are pushing the chat model to just chat and nothing else and want us to use thinking, which is not out yet, for everything else
last lotus
is this somehow what their massive normie customer base actually likes?
a listicles ai?
frozen pendant
Jesus this model is so bad
frozen pendant
is this somehow what their massive normie customer base actually likes?
You mean the gooners?
Idk if the gooners will like this either, seems like it regressed on writing
I’m not sure if gooning involves reading nice literature or only thing which matters is the gooning
brave spade
Codex is better at gooning than this
last lotus
You mean the gooners?
no no, not the gooners. obviously we know that there are two types of AI users: wise coders and degenerate gooners.
but there's also another kind, so often forgotten - yet they are by far the most numerous. the common person. the "regular folk". the average joes. End Users.
i don't think they're really used to reading that much, so you have deliver the information in a way they can understand. and maybe.... this is what's best for them
frozen pendant
no no, not the gooners. obviously we know that there are two types of AI users: ...
Regular users don’t pay tho
frozen pendant
Oh yes
I forgot there’s ads in the US
It does make sense actually
A friend of mine yesterday showed me he goes to the website on his phone and he doesn’t even log in to use ChatGPT
Absolute madness which 99% of the average users do too
last lotus
besides, this is AI we're talking about. it's not about making money. money's not real
someone seems to paying. i'm sure its fine
OpenAI’s own February 27, 2026 update states that ChatGPT has “more than 900M weekly active users” and “more than 50 million consumer subscribers.”
holy moley! maybe they do have to cut costs in some way
by making the model be cringe
frozen pendant
Advertisements will make it break even
And 10B slop models will make it profitable
glacial marsh
proxy test. at least not the worst (other than blind legal), but quite verbose+slow compared to other instants (aka -chats).
last lotus
signal wharf
https://fixupx.com/lefthanddraft/status/2029091899618041980
And above it at least 11 paragraphs or sentences on how to work with ads. What a joke
dusty tundra
https://fixupx.com/lefthanddraft/status/2029091899618041980
wow that is insanely cringe
pretty funny dichotomy
pseudo walrus
https://fixupx.com/lefthanddraft/status/2029091899618041980
they suck so much, they have to write prompt so the model wouldn't behave in certain way instead of tuning it
and oh, you must use emojis~
last lotus
🚨 PROMPT WATCH UPDATE https://fixupx.com/lefthanddraft/status/2033947454656356736
