OpenAI Counters DeepSeek with GPT‑3.5 Mini
Key Points
- OpenAI is slated to launch “GPT‑3.5 Mini” (referred to as 03 Mini) today around 10 a.m. PT, positioning it as a high‑performance, free‑tier option to counter DeepSeek’s competitive offering.
- The new model is expected to be faster and smarter than GPT‑3, giving free‑tier users up to 100 daily messages, which could pressure DeepSeek’s free tier and force OpenAI to reassess its pricing and packaging strategy.
- Sam Altman is reportedly in Washington, D.C., for a briefing that hints at longer‑term plans—potentially a GPT‑4 rollout in the first half of the year—highlighting the national‑security relevance of the technology.
- OpenAI continues to avoid public model naming conventions, making it unclear how “03 Mini” differs technically from the main GPT‑3 model beyond its smaller size and speed improvements.
Sections
Full Transcript
# OpenAI Counters DeepSeek with GPT‑3.5 Mini **Source:** [https://www.youtube.com/watch?v=379Isc7tJc4](https://www.youtube.com/watch?v=379Isc7tJc4) **Duration:** 00:06:31 ## Summary - OpenAI is slated to launch “GPT‑3.5 Mini” (referred to as 03 Mini) today around 10 a.m. PT, positioning it as a high‑performance, free‑tier option to counter DeepSeek’s competitive offering. - The new model is expected to be faster and smarter than GPT‑3, giving free‑tier users up to 100 daily messages, which could pressure DeepSeek’s free tier and force OpenAI to reassess its pricing and packaging strategy. - Sam Altman is reportedly in Washington, D.C., for a briefing that hints at longer‑term plans—potentially a GPT‑4 rollout in the first half of the year—highlighting the national‑security relevance of the technology. - OpenAI continues to avoid public model naming conventions, making it unclear how “03 Mini” differs technically from the main GPT‑3 model beyond its smaller size and speed improvements. ## Sections - [00:00:00](https://www.youtube.com/watch?v=379Isc7tJc4&t=0s) **Untitled Section** - ## Full Transcript
Chad GPT is firing back at Deep seek and
you know what you do when you're a
product team and these are hard models
to build and you can't just magically
build another model or change everything
all in a dime you respond with the thing
you originally planned but you position
it a little differently and they
originally planned open AI did to
release 03 Mini at the end of this month
the heavy heavy heavy rumor is that it
is today if they follow their usual
practice it will be at 10:00 a.m.
Pacific and about 4 half
hours and if they do that you will see
that 03 mini is going to be positioned
versus deep seek they probably won't
name deep seek on the call but it will
be
positioned as an amazing option even for
free users they've changed that it was
going to be available for plus users on
a very limited basis but deep seek has
upended the value chain with their free
tier and so going to push that reasoning
model down and even free tier users are
going to get it this is a model that is
supposed to be better than 01 so if
we're right if what Sam mman has
literally said is
true then free tier users will have
access to a better faster model with a
100 messages a
day and they still won't have a
one I don't know this is why they need
to get their business model and their
pricing and packaging figured out
because this is a
mess but the point is they're nervous
enough to change the availability for 03
mini they know they're going to burn
some cash as people use it but they need
to get that market share back and win
the Mind share of people who are running
to deep seek that's kind of that
strategy at the same time Sam has a very
busy day on what's next he is in
Washington DC uh I don't know how he's
getting there because Reagan is closed
which that's really bad um but he's in
DC doing a briefing and the briefing is
on rumor has it and this is this is
there are rumors that I feel good about
like 03 mini coming out in the next day
or so and there are rumors that I'm like
this is extremely sketchy but like
there's probably some information in
here somewhere this is in that latter
category it's extremely sketchy but
there's probably some information in
here somewhere the rumor is that it is
not 03 that he's in town to talk about
that's too close to production it's
something farther out but within the
first half of the year uh 04 has been
bandid around uh by some of the rumor
vultures that might be it the point is
that he feels like it has enough
National Security implications which
means Reed it's really really good that
it's worth going to Washington to talk
about one other tidbit on 03 again these
guys cannot name their models for the
life of them could you tell me the
difference between 03 mini and 03 I
would assume 03 mini is smaller as a
model but that doesn't explain why it's
faster
automatically anyway 03 mini is fast 03
mini is smarter than 01
apparently um and
03 was tested as part of the new
international AI safety report and the
results are just
bonkers the test results got submitted
and
it's O 01 has changed the game in the
fall I know it seems a long way away and
there are people who are like oh R1 is
better than o1 which for the record I've
played with them both I don't think it
is um but whether even if you believe
that 03 is just Bonkers good in the test
results and it's it's not 03 mini
they've got to pick a different name or
something they won't but they they ought
to when we get our hands on that it
looks like it's a model that is going to
be better than all but the top few
experts in the world at a wide range of
fields
and even 01 Pro isn't testing as well
it's it's just remarkable like I'm
really excited for 03 and 03 Pro so we
will see that is down the road as far as
I know they're not releasing 03 and 03
Pro today I believe it is only 03 mini
because 03 mini is a smaller model it's
easier to complete safety training and
yes it is a smaller model
and that is
that is one of those things that I wish
that they would think about less to be
honest with you because from a consumer
perspective I don't care how much
imaginary space your model takes up what
I care about is the capabilities of the
model what what can it actually do for
me what can it do now that it couldn't
do before and we have done a terrible
job explaining to people how that works
I actually wrote up a substack it's free
on the um on the substack talking about
how we don't understand the evals and
the acronyms like ime and uh GP QA and
like I did like what are these tests you
have this little like no did this and
this on the GP QA Diamond test and I'm
like what what is that mean so I went
and looked at the tests and I
screenshotted them and I wrote about
them and I kind of went through all of
them I even looked at Humanity's last
exam which just came out I think it's
worth taking a look at because we need
to understand as humans how we're
measuring these things and understand
how we're not measuring them like if I
looked at it and I did there's a lot of
math there's some real world
coding there's some literature stuff on
Humanity's last exam which they do the
worst
on
but by and
large we have no tests for the things we
tend to use AI for every day and I think
there is a gap there and I don't know
how to fix it
but I think that if we had a test that
was more understandable to people it
might help them figure out what AI can
do and how AI performance differences
are meaningful so something to think
about I don't think we've solved that
one but I think it's part of the problem
it's part of why people don't understand
what AI does and when a new model like
deep deep sea comes along they're like
what is this this is amazing or this is
terrible but it's mostly like their
priors their prior assumptions shaping
that rather than just sort of rigorously
thinking about what it can do y