What is the impact of DeepSeek on the AI sector? đŸ”„

The open source reasoner models from DeepSeek look not bad, I am impressed with their results and has a performance on par with OpenAI o1, So, could we expect a reduction in the price of tokens from OpenAI?

These are its prices

8 Likes

This is an interesting discussion thread:

https://www.reddit.com/r/LocalLLaMA/comments/1i5piy1/deepseek_r1_219m_tok_output_vs_o1_60m_tok_insane/

1 Like

Pure opinion but doubtful.

Open source has constantly shown that it’s right behind despite the walls put up. Being able to see the reasoning tokens is huge.

The OAI reasoning models seem to be more focused on achieving AGI/ASI/whatever and the pricing is secondary.


I’m curious to know how your experience has been with Deepseek. I’ve run coding tasks alongside gpt-4o, o1, and o1-mini and so far Deepseek isn’t shining as much as the benchmarks indicate.

That being said, the potential to use it’s data for training smaller models is huge. I got nothing but respect for the devs.

6 Likes

I doubt it but it would certainly be welcome. From a privacy perspective, the seeds of doubt were sown long ago and it would require some serious risk taking for many to jump ship to deepseek.

Happy to do it myself for my own use and experimentation :grin:

2 Likes

I like David Shapiro’s analysis about this topic:

3 Likes

Good question! The OpenAI API is indeed quite expensive. If competitors like DeepSeek continue to deliver similar performance with open-source models, there might be pressure on OpenAI to lower token prices to stay competitive.

2 Likes

The Deepseek impact:

3 Likes

Now we can see its impact on financial markets:

I think that OpenAI needs a more effective answer to this. Time is running out. Maybe OpenAI could decide to use the DeepSeek paper/model to improve o1, o3.

What could be the effect on StarGate?

4 Likes

Something weird is happening here. DeepSeek should accelerate proliferation. This is either a great buying op or there is some hardware competition that is behind this.

@sashirestela

1 Like

I can’t trust Sam because for him, ‘tons’ = maximum 100. :confused:

For me, 1 ton = 1000 (we use the metric system, not jokes :rofl:). If he says ‘tons,’ it should be at least 2000. That’s one thing.

The second thing is → I would use OpenAI ONLY if *Plus Subscribers get:

  • *Unlimited access to the o1 model and o1-mini.
  • *Limited access to the o3 and o3-mini.
  • *No silent updates → it’s disrespectful to users when they “tweak some parameters” and make models worse just to save on computation.
  • Prices equal to or comparable to Chinese models (for the API, or close if they add higher context).
  • For me, as I believe agents will be the future, I need a higher context for assistant instructions and functions.

Even with all that, I’m still not sure if it’s worth coming back



One week ago, I was thinking OpenAI was behind DeepSeek.

While I was researching them, I remembered Kai-Fu Lee talking about the Chinese in a video from a year ago → he said they would be so mad about taking data and offering the AI for free just to get the data.

I understand there’s a war over this technology, but making the model open-source → what kind of move is that? What’s really going on?
Are they ahead of the Americans and just trying to stop them from gathering data?

It’s a gambit here, like in chess → I think this is just the beginning.


This is normal; the price will rise again, and I think it will be above $150 at the end of the year → after Agents rise.

2 Likes

Yeh, you’re probably right, I tend to overestimate Mr Market. But for fun, let’s revisit this every week or so in this thread and see how it plays out.

2 Likes

I would expect an impact. Not just because R1 is actually not bad at all, but also because people now realize they can run LLMs locally.

For instance, just to try it out I installed Deepseek (and some other LLM models) on my own PC. Crated a simple Flask Python app that basically can handle incoming API calls (yes, it has authorization) with a prompt, then triggers a LLM and respond back. I just tested it and it works quite well. Basically I can now do my chat completion calls for free, even from my online apps.

Of course that will not work if many people use it at the same time, but - for instance - for nightly runs that make scheduled calls every sec or so it can work quite well


4 Likes

I tried DeepSeek vs chatgpt 4o 
 .I am really impressed with the results from DeepSeek. I gave same context to DeepSeek and Chatgpt to help me create an AI app. SeepSeek did it much better. Chatgpt kept getting stuck and producing code snippets with deprecated openai api(s). Aweful actually.

To early to make a call, but I am impressed.

2 Likes

I’ve watched a few videos on this subject, and actually subscribe to DeepSeek. In one video, the narrator asks the question: “Why would someone pay OpenAI, Google and Anthropic more for the same AI that DeepSeek provides?” My answer: For reliability. Look at DeepSeek’s status monitor over the past few days: https://status.deepseek.com/

I don’t know about anybody else, but I use AI to do text analysis on fairly large and complex documents. An actual business application. DeepSeek not only times out on the same inputs to which o1, Gemini and Claude easily respond, but it doesn’t even tell you it’s timing out. It simply doesn’t respond.

As others here have noted, it can be useful. But not something I think you want to rely on for enterprise applications, or your self-driving car or your humanoid robot. At least so far.

Note: I am referencing the API only.

6 Likes

This is mainly a personal opinion, but I think from Deepseek, it will help to reduce the pricing of AI all over the world to make sure pricing is competitive, so for developers at the end of the day it is a + where we get better AI for lower prices, and I think now that OpenAI has a proper competitor it will lead to more and more innovation and would result in a better AI sector.
I don’t think 1 will win at this point because there is a lot to see on what will happen but this will be a historic moment in the history of AI.

4 Likes

From today’s newsletter from The Information:

There are also questions about the way its owner, High-Flyer Capital Management, achieved its results (as we detail here).

The link then leads to Meta’s reaction to the R1 release.

Can maybe anyone with a subscription share a summary of what is being discussed?

4 Likes

I think DeepSeek might be less stable than his more established competitors, but it’s something that could be quick fixed given his popularity. Anyway total dominance of one country in AI is a very very dangerous thing for humanity - especially when the whole power is concentrated in a hands of very few people. I want to see future when AI system is like a local app and you need a cloud only for very specific hardcore tasks, so most of your private data stays on your computer.

6 Likes

Interesting 
 I checked the update today using the url you provided. Here is the status on 1/28/2025:

" Due to large-scale malicious attacks on DeepSeek’s services, we are temporarily limiting registrations to ensure continued service. Existing users can log in as usual. Thanks for your understanding and support.
Jan 28, 2025 - 17:07 CST"

2 Likes

I do hope this is a temporary situation as I would love to be able to incorporate the API into my application. It just bothers me to see all these YouTube videos predicting the demise of OpenAI and Google because of the advent of “cheap” AI, ignoring the fact that “cheap” doesn’t always only mean only “low cost”.

We pay a premium price to these big companies in exchange for predictable, consistent and reliable service.

3 Likes

they stole gpt-4 a while ago, they won’t be able to keep up. It’s all good, just wait for the newer models to drop.

2 Likes