Key Points
- DeepSeek launches two new AI models: V3.2 and V3.2‑Speciale.
- V3.2 is available through DeepSeek’s app and web platform.
- V3.2‑Speciale is accessed via API and targets high‑end reasoning tasks.
- Company claims V3.2‑Speciale outperforms Google’s Gemini 3.0 Pro and OpenAI’s GPT‑5 High in benchmarks.
- Models feature 685 billion parameters, requiring extensive server hardware.
- DeepSeek stresses an open‑source approach, allowing free model loading for other companies.
- The models aim to challenge leading AI providers and expand DeepSeek’s market presence.
New Models Expand DeepSeek’s Portfolio
DeepSeek introduced two advanced AI models named V3.2 and V3.2‑Speciale. V3.2 follows the earlier experimental release V3.2‑Exp and is marketed as a balanced solution for everyday use. The model is currently accessible through DeepSeek’s mobile app and web interface, allowing users to experience its capabilities without specialized hardware.
V3.2‑Speciale Targets High‑End Reasoning Tasks
The Speciale variant is restricted to API access, positioning it for developers and enterprises that need powerful reasoning functions. According to DeepSeek, V3.2‑Speciale outperforms Google’s Gemini 3.0 Pro and OpenAI’s GPT‑5 High in a series of benchmark evaluations, though the company did not publish a direct comparison with GPT‑5 Pro.
Performance Claims and Competitive Positioning
DeepSeek asserts that V3.2 reaches reasoning performance “on par” with Gemini 3.0 Pro, while V3.2‑Speciale achieves “gold‑level” results in competitive settings such as the International Mathematical Olympiad and the International Olympiad in Informatics. The company frames these achievements as evidence that it can rival the leading Western AI firms.
Open‑Source Commitment and Infrastructure Requirements
DeepSeek emphasizes its commitment to open‑source principles, stating that its models can be loaded by any company free of charge. However, the V3.2 model contains 685 billion parameters, necessitating a large server infrastructure and substantial investment in hardware to run effectively.
Industry Reaction and Context
The release comes as DeepSeek continues to push the narrative that high‑performance AI models can be offered at lower cost than competitors. The company’s earlier free release of a reasoning‑capable model garnered attention for challenging the development pace of firms like OpenAI, Google, and Anthropic. DeepSeek’s latest models reinforce its strategy of providing powerful AI tools while maintaining an open‑source stance.
Source: cnet.com