DeepSeek has announced the launch of the DeepSeek-R1 model, which is considered a qualitative leap in the field of open source artificial intelligence. This model offers very high performance that competes with the O1 model from OpenAI, and is distinguished by its efficiency in dealing with mathematical problems, programming, and solving complex problems. The model includes 671 billion parameters, but only 37 billion are used during operation, which makes it powerful and efficient in terms of resource consumption. It also uses reinforcement learning techniques in the post-training stage, which helps it achieve exceptional performance with very limited data. The most important feature of DeepSeek-R1 is that it is open source under the MIT license, which means that it is available for commercial use and modification freely, which encourages innovation and collaboration between developers and companies. In addition to the basic model, DeepSeek has launched 6 lighter versions ranging from 1.5 billion to 70 billion parameters, and the 32B and 70B versions have proven that they are able to compete with the O1-mini model from OpenAI. The model is available for use via the chat.deepseek.com website, which offers a “Deep Thinking” mode, or via the developer API at very competitive prices, such as $0.14 per million input tokens and $2.19 per million output tokens. Global benchmarks such as AIME, MATH-500, and SWE-bench Verified have confirmed that DeepSeek-R1 excels in logical reasoning, deep analysis, and answer verification. This model is not only a tool for developers, but also a big step forward for companies looking for powerful and flexible AI solutions.

Yes please , this will reduce cost of operation greatly

@orange-egg-72655 👀 (trying to get this post seen ahah) In the meantine i think i'll start to build my own integration with deepseek api and share it with you guys to be continued...

Got it! I'll let the core engineering team know ;D

great to hear, thanks a ton !

Yeesssss

Hey everyone!! The DeepSeek R1 and V3 models are now available in bots through the __Fireworks.ai__ integration. If you already have a bot with the integration installed, then just update it to the latest version to get access to it. For future reference: we are looking into supporting v3 and r1 through other providers. However, for now, we have at least something to show that works!!!! For those wondering, there are concerns in the AI community about using Deepseek's api directly because of their terms and conditions (specifically section 4.3 of their terms and services which explicitly states that they will use inputs and outputs to train their AI, which raises concerns of security of clients) --> Which is why we use a third party currently, fireworks.ai, instead of deepseek's api directly. However, it's up and running!! Feel free to leave your questions and concerns in the help chat / here. I look forward to hearing what you, the community, has to say about it! Happy bot building :DD

thanks Matthew !! i saw that yesterday, that was quick ! i understand the concerns you have about deepseek's terms and conditions, that's indeed something to worry about The only thing i have to say right now is the price of deepseek models through fireworks, which is soooo much higher than the real prices for these models (i think they charge $1/$1 for v3 and $8/$8 for r1 when it's 0.44$/$2.19 for r1 via deepseek api approximately) As you plan to bring the possibility to integrate it with other providers, that could be resolved Again, thanks for the reactivity 🙂

Yes, I agree with people like me, what matters at this stage is controlling costs and reducing them as much as possible with high-output language models.

I've brought up your concerns to the team! We will be keeping everyone updated as to new updates/findings regarding deepseek :DD

You're the best 😍