So they made their model "smaller" to rival Deepseek, which was 20 times as efficient than there larger model? That seems like a PR stunt.Imagine Apple brought an "iPhone light" that uses cheaper components to compete with the price of a Chinese phone.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
OpenAI Launches o3-mini, a Cost-Efficient Reasoning Model Rivaling DeepSeek
- Thread starter MacRumors
- Start date
- Sort by reaction score
I don't think the case is so clear-cut for OpenAI. Their operating costs are immense, anyone using GPT costs them money. Only about 5% of users pay subscriptions, and even the $200/month pro plan is not profitable. The company runs an annual loss of 5 billion dollars, and they were on the verge of bankruptcy in late 2024. Big user numbers can impress investors for a while, but eventually they will want to see this translate into paying customers and profits, and OpenAI does not seem to know how to get there.
The investors who saved OpenAI do not necessarily believe in a future for OpenAI: Nvidia wants to keep the AI hype going in order to sell GPUs, and OpenAI collapsing would damage the hype. Microsoft has effectively owned OpenAI for years, and OpenAI dying would make them look bad. Meanwhile they struggle to get some profit from this investment, as users do not want to pay for the OpenAI-based Copilot shoehorned into Windows and Office.
Depends what you mean by consumer hardware. 5 year old second hand servers fully populated with ram? Sure you can.
Mac Pro 2019 with 1.5tb of ram? Yup!
I'm running the Deepseek Distilled locally on an M1 Max MBP. It takes about 26gb of ram. So it's not the 300gb+ version, but results are pretty close.
More importantly, Deepseek showed that powerful, modern models can be made that can run locally. This is good for everyone and for privacy long term. It's especially good for Apple as they have been pushing towards running models at the edge.
Running an 8-bit quantized version of deepseek-r1-distill-llama-70b through LM Studio:
That is just one of the more interesting attempts to get information from the model installed locally on my Mac. Different attempts will give different answers. I wouldn't doubt that with the right kind of prompting you can get some information out of it. But, the fact is, DeepSeek is post processed to adhere to guidelines set by the Communist Chinese Government.
Last edited:
One company trying to just own the collective creation of humanity is absurd.
How exactly should open ai be able to take the entirety of human creation, all text that has been written. And claim it their own? How do you square that circle? AND call their company OPEN ai? While they vacuum ALL text and images available to make an Ai...
No, it HAS to be open source.
The biggest revelation is not that open-ai level Ai is doable by other sources. It`s the fact that IT DOESNT TAKE A SUPERCOMPUTER TO RUN IT!!!
Ai is the 1st program that a company is trying to own in its entirety. No competitors allowed..
AND open ai has the gall to say deepseek copied them... No, they build on-top of human knowledge, ai was not invented by open ai. They just tried to monopolize it and failed.
How exactly should open ai be able to take the entirety of human creation, all text that has been written. And claim it their own? How do you square that circle? AND call their company OPEN ai? While they vacuum ALL text and images available to make an Ai...
No, it HAS to be open source.
The biggest revelation is not that open-ai level Ai is doable by other sources. It`s the fact that IT DOESNT TAKE A SUPERCOMPUTER TO RUN IT!!!
Ai is the 1st program that a company is trying to own in its entirety. No competitors allowed..
AND open ai has the gall to say deepseek copied them... No, they build on-top of human knowledge, ai was not invented by open ai. They just tried to monopolize it and failed.
Simple, as everything on the internet is free, unless of course it is what you created and then it's mine and mine alone.
-- The for profit AI company motto
The only thing that doesn't suck is the vacuum...
They're trying to find someway to make a profit, and probably being called 'open" is a plus.
I use their service right now because on a pay as you go basis it's good enough for prototyping AI features and faster than running in on my M3 Mac. If we decide to do it locally we'll invest in a machine with huge RAM and probably run Linux on a non Mac box for cost reasons. That, however, then means running a server with teh associated costs as well; vs hosting elsewhere and using the APIs.
I find OpenAI more reliable than Deep Seek as DS sometimes comes up with weird results using the same test prompts and data sets.
Still not open source… I’m good with DeepSeek bud 👍
OpenAI today released o3-mini, a new, more cost efficient model that's available for both ChatGPT and the OpenAI API starting today. OpenAI previewed o3-mini in December, but it is ready at a time when OpenAI is aiming to keep up with Chinese company DeepSeek.
According to OpenAI, o3-mini "advances the boundaries" of what's possible with smaller, more efficient models. o3-mini is designed for STEM capabilities, and OpenAI says that it has particular strength in science, math, and coding. o3-mini is search capable, and it can locate up-to-date answers with links to relevant web sources.
o3-mini is OpenAI's first small reasoning model that supports features like function calling, Structured Outputs, and developer messages. There are three reasoning effort options, including low, medium, and high, to optimize for different use cases that prioritize processing power for complex challenges or speed.
OpenAI is keeping o1 as its standard general knowledge reasoning model, but o3-mini is available for technical domains that require precision and speed. When used with ChatGPT, o3-mini uses medium reasoning effort for a balance between speed and accuracy, but o3-mini high is an option in the model picker.
ChatGPT Plus, Team, and Pro users are able to use o3-mini starting today, and it will replace o1-mini in the model picker. The rate limit for Plus and Team users is now 150 messages per day, up from 50.
o3-mini is the first reasoning model that OpenAI is making available to free users. Those with the free plan can try o3-mini by selecting the "Reason" option in the message composer.
Article Link: OpenAI Launches o3-mini, a Cost-Efficient Reasoning Model Rivaling DeepSeek
I wonder how long it takes for a downloaded AI system to upload all your personal information to China or a compatriot? Copying does not mean your data was removed so no trace of that theft may remain. Sleep well until your bank starts bouncing checks as there are no funds left in your account....
That would depend on what you downloaded. People downloading DeepSeek are not downloading an app. They are downloading parameters and loading those into a (hopefully) trustworthy execution app, like Private LLM, or LM Studio. So, the chance of China getting your personal data because you downloaded the DeepSeek model are nil.
Hopefully, one could "read" the source code's every line to determine the veracity of your statement. An external call could load the data theft program and it would delete itself when done....
The models are just matrices of numbers. There are no external calls in them. That doesn't mean the app into which you load the model is free from trojans. The source code for LM Studio is available on GitHub for your perusal.
Last edited:
"Rivaling DeepSeek" and "aiming to keep up with Chinese company DeepSeek" were truly 😬
Not surprisingly, main stream media did not understand what DeepSeek is all about and cooked a compelling storyline that was simply incorrect. This astonishingly put markets in chaos.
How about MacRumors do better and learn what is really happening in AI (and DeepSeek). Dario Amodei's blog "On DeepSeek and Export Controls" is a good place to start.
Last edited:
You might have noticed that o3-mini was introduced in Dec 2024. It is not a reaction to DeepSeek R1, but other way around.
With statements like this, please specify which models you compare: There are multiple variants for o1, o3, and DeepSeek. They are all very different in their performance.
Drive towards $0.5T datacenter investment is a plan for being the first with capacity to host super-intelligent AI agents in scale that will replace large share of the workforce globally. While an extremely bold bet, the lead it could give US over China may be pivotal as it is possible that this "datacenter full with geniuses" leads to further acceleration that keeps rivals behind.
If I would be EU I would be terrified. But EU seems clueless.
China seems to understand this and is taking steps to mitigate. But there are limits — DeepSeek was merely able to build a ~ $1B datacenter. Export controls are hurting them. I fear all this will lead to a hot war with Taiwan.
Yep. You are right. I've been using an 8-bit quantized deepseek-r1-distill-llama-70b. I get that I am not testing the full 1.34 TB BF16 version of DeepSeek V3. If I could buy an Apple Silicon Mac Pro that would take 2TB, I'd be running unreduced models.