Remember DeepSeek, the large language model (LLM) out of China that was released for free earlier this year and upended the AI industry? Without the funding and infrastructure of leaders in the space like Google and OpenAI, the team behind the DeepSeek R1 model was able to create a viable competitor to Gemini and ChatGPT.
This week, the DeepSeek team released two powerful new models, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale — and once again they are totally free to use.
This Tweet is currently unavailable. It might be loading or has been removed.
DeepSeek’s China-based team says that DeepSeek-V3.2 provides equivalent levels of performance as OpenAI’s GPT-5 and described the model as “your daily driver.” DeepSeek-V3.2 is more for the same everyday tasks that you’d seek out from a AI assistant chatbot like ChatGPT.
DeepSeek-V3.2-Speciale, on the other hand, appears to be competing with the more advanced AI models like Google’s Gemini 3.0 Pro. DeepSeek describes the model as having “maxed-out reasoning capabilities” and shared testing results showing DeepSeek-V3.2-Speciale acing numerous LLM reasoning benchmarks.
“DeepSeek-V3.2 is our first model to integrate thinking directly into tool-use, and also supports tool-use in both thinking and non-thinking modes,” DeepSeek said in a statement announcing the models posted to its X account.
Basically, as the DeepSeek team describes it, the latest model is able to mimic human reasoning and utilize tools like search engines and calculators. Furthermore, DeepSeek-V3.2-Speciale is for mathematical needs, with the team sharing the models’ “gold medal” results from taking tests such as the International Math Olympiad.
Mashable Light Speed
DeepSeek says it is setting out to “push the inference capabilities of open-source models to their limits and explore the boundaries of model capabilities.”
How does DeepSeek-V3.2 compare to GPT-5?
The biggest difference is that DeepSeek’s models are open-source. In short, they’re tools that can be applied to specific problems. ChatGPT, on the other hand, is a closed-source, general-purpose AI chatbot that combines multiple models into a single user experience.
So, while it seems that the latest DeepSeek models stack up to ChatGPT and other models contextually, it should also be noted that it is primarily a text-based model and not one for images, video, or other media. And, of course, DeepSeek’s models are free, while users have a very limited number of prompts with the free version of ChatGPT.
So far, on the popular AI leaderboard LMArena, the new DeepSeek models have lagged behind the likes of GPT-5 and Gemini 3. And in terms of feature set, ChatGPT still offers a more complete, all-around experience.
How can you try DeepSeek-V3.2?
Users currently looking to utilize DeepSeek-V3.2 can do so right now on DeepSeek’s website, API, and app. You can head to the DeepSeek platform and try it right away.
However, as of now, the DeepSeek-V3.2-Speciale model is only available through DeepSeek’s API. The API access for DeepSeek-V3.2-Speciale does incur a cost; however, it is 10 times cheaper than its competitors’ models, with a cost of 42 cents per million tokens.
DeepSeek last upended the industry nearly a year ago. Since then, a lot of advances have been made with other models. We’ll soon see if DeepSeek’s latest models continue to disrupt the industry.
Topics
Artificial Intelligence
DeepSeek