Access to all API's except gpt-4-32k: Why?

I have access to the ‘gpt-4’, ‘gpt-4-1106-preview’, and ‘gpt-4-1106-vision-preview’, API’s. But when I try ‘gpt-4-32k’, I get the error:

Error code: 404 - {‘error’: {‘message’: ‘The model gpt-4-32k does not exist or you do not have access to it. Learn more:’, ‘type’: ‘invalid_request_error’, ‘param’: None, ‘code’: ‘model_not_found’}}


It’s because gpt-4-32k had a very limited rollout.

So it’s not you, it’s just that limited, even to this day.

But the latest GPT-4 variant, gpt-4-1106-preview, is 128k. So use that instead if large context is required.


I think that, for my uses, gpt-4 produces a superior result to gpt-4-1106-preview. That is fine for now, but is there any waitlist I can sign up to for gpt-4-32k long term?

I haven’t heard of any recent waitlist activity for 32k. It used to be you submit evals.

But in light of the new 128k model, the demand for 32k likely has evaporated.

Plus 32k is 4-5x more expensive than the new 128k version.

So I’ve moved away from 32k due to cost alone.

Ok, thank makes perfect sense, thank you.
I am fine with using 128k then.
But, what are “submit evals”?

From the OpenAI GitHub:

Evals is a framework for evaluating LLMs and LLM systems, and an open-source registry of benchmarks.

Evals provide a framework for evaluating large language models (LLMs) or systems built using LLMs. We offer an existing registry of evals to test different dimensions of OpenAI models and the ability to write your own custom evals for use cases you care about. You can also use your data to build private evals which represent the common LLMs patterns in your workflow without exposing any of that data publicly.

1 Like

In my opinion, gpt-4-32k-0314 used to be OpenAI’s most capable model. 0613 was a downgrade as far as I’m concerned, but I think that is what GPT-4 currently links to. In most use-cases, gpt-4-1106-preview seems to slightly outperform 0314 on top of having a more recent corpus (and being way cheaper, and considerably faster).

But it could really depend on what exactly you’re trying to do.

Should we take a look at your prompts so you might be able to get better results with the available models?


GPT-4-32k was also the fastest lowest latency GPT-4 model too.

But I am not using it anymore, so I don’t have any recent latency data, but from what I remember it was around 2x faster than the vanilla 8k GPT-4.

But this was all “early days” stuff.

To put things in perspective:

Latencies I have on record are:

[Output Tokens Per Second = OTPS, includes round trip delays to the API and back]
gpt-4-1106-preview: 30 OTPS
gpt-4-32k: 20 OTPS
gpt-4 (8k): 10 OTPS

But literally 1 day ago, they sped up vanilla gpt-4 to match the old gpt-4-32k at 20 OTPS. :rofl:

So these latencies are highly transient and can change on a dime.

system_content = f'''You are a pro YouTuber. You specialize in writing fluid video essays, section by section. You write scripts that are ready to read without any changes. Never include cues, notes, instructions, or reminders. only script text (!!). You are charismatic.'''
message_content = f"""Write me an emotionally engaging, data-driven, quick, short-as-possible introduction, that has a hook, sets the stage, and incentives the viewer to watch until the end. Here is the outline: '''{outline}'''. Never include cues, notes, instructions, or reminders. Return only the script."""

It is kind of hard for me to tell if gpt-4 or gpt-4-1106-preview produces better results, since the quality of the script is subjective. I think the script is more fluid with gpt-4, but, again, subjective.
Price is not much of a concern with me. I am willing to go with the best model for (1) fluidity and (2) accuracy (meaning there shouldn’t be any misinformation).

Which of the available models do you think would be best?

Is there any way to access that old model, ‘gpt-4-32k-0314’?

super hard to say. I mostly evaluate on reasoning performance, don’t care too much about style. maybe @PaulBellow might be highly qualified to answer this :thinking:


Imagine the heart-pumping action amplified by the roaring engines of motocross bikes as the world’s top riders gather in one place—welcome to the Red Bull Motocross Championships in Los Angeles! Get ready for an unparalleled deep dive into the world of extreme motocross, where we’ll explore what it takes to become a true champion.

Join us as we uncover the incredible stories and achievements of the riders who make this event truly spectacular. Discover the fierce dedication that drives these athletes to push themselves beyond their limits and overcome countless obstacles on their quest for glory.

Go behind the scenes to witness the intricate and precise preparations that are essential to this exhilarating event: from track setup to rider practice sessions and even insider interviews with the unsung heroes that make it all possible—the team crew members.

We’ll take you into the heart of the pulse-quickening battle as riders defy gravity and fly across the track. With exclusive drone footage and slow-motion shots, you’ll feel like you are in the center of the high-octane action. And, of course, we’ll capture the game-changing moments that keep fans on the edge of their seats.

After the dust has settled, hear from the victors themselves as they reflect on their performance. Were their strategies the winning ticket, or did fate lend a helpful hand? And what about the other competitors, where do they go from here as they prepare to face their next challenge?

As we bid farewell to the adrenaline-fueled world of Red Bull Motocross Championships, we’ll leave you with a taste of what’s next in store. So buckle up and prepare for the ride of your life—this is one show you won’t want to miss!


Imagine the roar of engines vibrating through your body, the scent of burning rubber in the air, and the taste of adrenaline on your tongue - welcome to the Red Bull Motocross Championships in Los Angeles, the peak of high-octane thrill.

Over the years, this event has been decorated with unprecedented stunts, heart-stopping showdowns, and nail-biting finishes. And today, we’re bringing you a whole level closer. Beyond your screen, to the heart of the track, where fearless riders conquer gravity and obstacles alike to emerge as victors.

You’ll meet our key contenders - each making a name for themselves, not by playing it safe but by challenging the odds. Get a glimpse of their unyielding spirit, the grit behind their inspiring journeys, and the moments of their glory that got them here.

Then, join us as we head behind the scenes. Here, every second counts. Uncover the chaos and precision that leads up to the main event. Witness the meticulous preparation, the practice sessions, and engage with the unsung heroes of the crew.

Hold your breath as we dive into the heated action. The tension on the start line, the exhilaration of the jumps, the spectacles in slow motion, and the inescapable drama unfolding from our bird’s eye view. Relive the thrilling twists, turns, the heartbreak of crashes, and the glory of recoveries that is Red Bull Motocross.

A taste of victory is sweeter when shared. Sit with us, as our victors reflect on their journeys, their victories, and their formidable competitors. Hear their experiences, learn their insights, and perhaps, catch a glimpse of their future endeavors.

As we wrap up, we don’t just leave you with a recap. We leave you ready to anticipate, to speculate on our next venture - a story that’s yet to be written. Because with Motocross, the thrill never ends. So, until next time, keep the engine of excitement revving in your hearts. This is Red Bull Motocross Championships signing off. Stay fierce, stay relentless.


Dirt flying, engines roaring, and hearts racing—the unmistakable cocktail of a Red Bull Motocross Championship. And this time, the battleground is none other than the City of Angels, Los Angeles. Welcome, thrill-seekers, to a spectacle where gravity is defied, legends are born, and the smell of victory is stronger than exhaust fumes.

We’re not just talking about any race; we’re talking about a symphony of man and machine that has left crowds gasping and competitors hungry for glory. Here, in previous years, the impossible was made possible, where riders became airborne artists painting their bravado across the sky.

Today, it’s not just about the bikes that thunder beneath them; it’s about the warriors who mount them. You’ll get exclusive, up-close-and-personal encounters with the riders who have etched their names into the dirt—heroes who’ve turned the impossible into I-just-did-that. Through sweat-stained interviews and heart-thumping highlight reels, we’ll dive into what drives them towards the finish line.

And what’s an epic without a glimpse behind the curtain? See the meticulous orchestration of track-setup, the raw nerve of pre-event jitters, and the symphony of sounds in a rider’s practice session. Get the real scoop from the pit crew who turn wrenches into wands, making motocross magic happen under the most intense pressure.

Hold tight as we unleash the pandemonium of the event itself. We’ll serve you high-flying jumps, gravity-defying whips, and daredevil passes that redefine the word ‘close-call.’ Witness the ballet of bikes in slow-mo, and feel your pulse match the RPMs of the beastly machines captured in exquisite drone footage.

When the dust settles, you’ll witness the raw, unfiltered emotions of the winners as they unpack their trials and triumphs. And let’s not forget the brave competitors who, despite not topping the podium, share insights only a racer knows and forecast their comebacks with fire in their eyes.

As we cross the finish line of our journey together, we’ll wrap up the heart-stopping moments and drop a teasing glimpse into the future, where the saga continues.

So, sit back, strap in, and prepare for a ride that’s more than just a race—it’s a testament to the human spirit, powered by Red Bull. Welcome to the ultimate adrenaline rush on two wheels. Welcome to the Red Bull Motocross Championships in Los Angeles.

user prompt

Write me an emotionally engaging, data-driven, quick, short-as-possible introduction, that has a hook, sets the stage, and incentives the viewer to watch until the end. Here is the outline: ‘’’

I. Introduction
A. Opening shot of Red Bull logo
B. Introduce the event and location (e.g., Red Bull Motocross Championships in Los Angeles)
C. Exciting clips from previous events

II. Rider Profiles
A. Introduce key riders competing in the event
1. Short interviews with each rider
2. Highlights of their previous accomplishments

III. Behind the Scenes
A. Glimpse into the pre-event preparations
1. Track setup and maintenance
2. Rider practice sessions and warmups
3. Interviews with team crew members

IV. Event Highlights
A. Showcase adrenaline-pumping race action
1. Jumps, whips, and passes
2. Rider close-ups and slow-motion shots
3. Drone footage of the race
B. Emphasize key moments throughout the event
1. Crashes and recoveries
2. Exciting photo finishes
3. Unexpected twists or surprises

V. Post-Event Interviews
A. Interviews with the winners
1. Discuss their experience throughout the event
2. Share their thoughts on their performance and competitors
B. Reactions from other competitors
1. Insights on their performance
2. Plans for future competitions

VI. Conclusion
A. Recap of event highlights and results
B. Teaser for the next event
C. Closing shot of Red Bull logo and outro music

‘’'. Never include cues, notes, instructions, or reminders. Return only the script.

huh it actually says “engine not found” for me on openai lol, but it’s still listed on my playground. On azure, it seems I can only create new deployments of regular gpt-4 0314 in france :thinking:, but 32k only 0613 everywhere else. It’s certainly on its way out :frowning:


I don’t believe 32k access is going to be released to anymore people now that the 128k model is now public. The fact that 32k performs better at some aspects than the 128k model is probably not taken into account by the OAI staff in charge of tell releasing the API models.

1 Like

It’s probably not worth dedicating the additional GPUs to make it acceptably available to more users because very few would actually use it.

It’s more than 4x as expensive as GPT-4 Turbo, any areas where it may perform better probably don’t hold when you consider the possibility of multiple passes with GPT-4 Turbo.

It’s just so expensive!

$120 / 1M Tokens generated.


OK, this is super helpful. I am not style expert either, but these comparisons help give me a better idea

1 Like

What aspects would you say 32k performs better at?

What do you mean by “multiple passes”? Setting n > 1?
(I’m still a bit of a novice)