In a bold move to reclaim its position in the global artificial intelligence race, China's DeepSeek has launched two powerful new AI models. The DeepSeek-V3.2 and DeepSeek-V3.2 Speciale are positioned as direct competitors to the reigning champions, OpenAI's GPT-5 and Google's Gemini series. This launch marks a significant attempt by the Chinese startup to disrupt the current dominance of Western AI giants in the chatbot and reasoning space.
DeepSeek's Comeback: Challenging Western AI Supremacy
The beginning of 2024 saw DeepSeek rise to prominence with its cost-effective DeepSeek R1 and V3 models, which delivered performance comparable to established players like ChatGPT. However, the spotlight later shifted to other Chinese firms like Qwen and Kimi, and Western advancements from Google's Gemini Nano series and OpenAI's agentic AI releases and GPT-5.1 upgrade. DeepSeek's latest release is a strategic effort to change this narrative and retake centre stage.
What Makes the DeepSeek-V3.2 Models Special?
The company describes the V3.2 series as 'reasoning-first models built for agents'. DeepSeek makes a strong claim that the standard V3.2 model delivers GPT-5 level performance in general tasks. The more advanced V3.2-Speciale model, however, is where the real competition heats up.
DeepSeek asserts that the Speciale variant offers reasoning abilities on par with Google's latest Gemini 3 Pro, particularly in complex problem-solving. The model's prowess is backed by impressive benchmark results:
- Achieved Gold-medal performance in the 2025 International Mathematical Olympiad (IMO) and International Olympiad in Informatics (IOI), a feat previously dominated by proprietary models from Google and OpenAI.
- Outperformed both GPT-5 High and Gemini 3 Pro on pure math benchmarks like AIME and HMMT.
- Led the pack in pure coding logic on the CodeForces benchmark, defeating all other listed AI models.
- On the notoriously difficult Humanity's Last Exam (HLE) benchmark, it scored higher than GPT-5 High but slightly behind Gemini 3 Pro.
How to Access the New DeepSeek AI Models
For regular users looking to experiment, the process is straightforward. The DeepSeek-V3.2 model has replaced the older V3.2-Experimental version (released in September) as the default on the DeepSeek chatbot website and mobile app. Users simply need to log in to start using it immediately.
However, access to the more powerful DeepSeek V3.2-Speciale is currently restricted to API use only. The company has not provided a timeline for when this high-performance model will be made available to the general public through its standard platforms.
The Implications for the Global AI Landscape
This launch signifies a renewed and potent challenge from China in the foundational model arena. By claiming parity with the best from OpenAI and Google in reasoning and mathematical tasks, DeepSeek is directly contesting the narrative of Western technological supremacy in AI. The move promises to intensify competition, potentially leading to faster innovation and more accessible, powerful AI tools for developers and users worldwide. The performance claims, if independently verified, could significantly alter market dynamics and user preferences in the coming months.
