Recall DeepSeek, the substantial language model (LLM) from China that was made available for free earlier this year and shook up the AI sector? In spite of its lack of funding and infrastructure compared to giants like Google and OpenAI, the crew responsible for the DeepSeek R1 model succeeded in developing a viable rival to Gemini and ChatGPT.
This week, the DeepSeek group introduced two powerful new models, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale — both offered at no cost.
The China-based DeepSeek team asserts that DeepSeek-V3.2 delivers performance levels on par with OpenAI’s GPT-5 and characterizes the model as “your daily driver.” DeepSeek-V3.2 is tailored for routine tasks typically managed by AI assistant chatbots like ChatGPT.
On the other hand, DeepSeek-V3.2-Speciale appears to focus on more sophisticated AI models such as Google’s Gemini 3.0 Pro. DeepSeek describes this model as boasting “maxed-out reasoning capabilities” and has released testing results demonstrating DeepSeek-V3.2-Speciale outperforming several LLM reasoning benchmarks.
“DeepSeek-V3.2 is our initial model to incorporate thinking directly into tool-use, and also facilitates tool-use in both thinking and non-thinking modes,” DeepSeek conveyed in an announcement shared on its X account.
Fundamentally, the latest model can replicate human reasoning and employ tools like search engines and calculators. Furthermore, DeepSeek-V3.2-Speciale is specifically designed for mathematical procedures, with the team revealing its “gold medal” accomplishments from tests like the International Math Olympiad.
DeepSeek aspires to “push the inference capabilities of open-source models to their limits and examine the boundaries of model capabilities.”
How does DeepSeek-V3.2 stack up against GPT-5?
The primary distinction is that DeepSeek’s models are open-source, which means they’re instruments that can be utilized for particular challenges. In contrast, ChatGPT is a closed-source, general-purpose AI chatbot that amalgamates various models into a cohesive user experience.
Although the latest DeepSeek models appear to be comparable to ChatGPT and other models contextually, it’s important to acknowledge that they are mainly text-based and not intended for images, video, or other media. Additionally, DeepSeek’s models are free, while the complimentary version of ChatGPT provides a very limited number of prompts.
On the popular AI leaderboard LMArena, the new DeepSeek models have so far fallen behind names like GPT-5 and Gemini 3. In terms of features, ChatGPT still offers a more comprehensive, all-around experience.
How can you experiment with DeepSeek-V3.2?
Users can currently access DeepSeek-V3.2 on DeepSeek’s website, API, and app. You can visit the DeepSeek platform and test it out right away.
However, the DeepSeek-V3.2-Speciale model is exclusively available via DeepSeek’s API. While accessing DeepSeek-V3.2-Speciale through the API entails a fee, it is ten times less expensive than competing models, priced at 42 cents per million tokens.
DeepSeek last made waves in the industry nearly a year ago. Since then, there have been considerable advancements with other models. We’ll soon find out if DeepSeek’s latest models keep disrupting the industry.