DeepSeek-V3.2
DeepSeek AI introduced DeepSeek-V3.2, a model that harmonizes high computational efficiency with superior reasoning and agent performance, on 1 December 2025. Their approach is built upon three key technical breakthroughs:
- DeepSeek Sparse Attention (DSA): An efficient attention mechanism that substantially reduces computational complexity while preserving model performance, specifically optimized for long-context scenarios.
- Scalable Reinforcement Learning Framework: A robust reinforcement learning protocol and scalable post-training compute allow DeepSeek-V3.2 to perform comparably to GPT-5.
- Large-Scale Agentic Task Synthesis Pipeline: Facilitates scalable agentic post-training, improving compliance and generalization in complex interactive environments.
The DeepSeek-V3.2 reasoning model was released on PolyU Gen AI on 1 January 2026 for use by PolyU staff and students, replacing DeepSeek-R1-0528.
Microsoft Foundry
The DeepSeek-V3.2 service, powered by Microsoft Foundry, will incur a usage fee of 0.58 credits per 1,000 tokens for input and 1.68 credits per 1,000 tokens for output. Only non-thinking mode is supported in DeepSeek-V3.2 hosted by Microsoft Foundry.
Alibaba Cloud Bailian
The DeepSeek-V3.2 service, powered by Alibaba Cloud Bailian, will incur a usage fee of 0.28 credits per 1,000 tokens for input and 0.42 credits per 1,000 tokens for output, plus the cost of chain-of-thought tokens. You may choose to enable / disable the thinking mode of DeepSeek-V3.2 hosted by Alibaba Cloud Bailian.
- Thinking Mode: In this mode, the model takes time to reason step-by-step before delivering the final answer. This is ideal for complex problems that require deeper thought.
- Non-Thinking Mode: Here, the model provides quick, near-instant responses, suitable for simpler questions where speed is more important than depth.
Please note that the reasoning steps of DeepSeek-V3.2 will also consume user credits, which may result in a further increase in the overall fee for each conversation. Due to its cost and longer reasoning time, use DeepSeek-V3.2 with thinking mode for tasks that demand advanced reasoning capabilities. For general tasks, using the model in non-thinking mode is often the better choice.
Staff and students may leverage reasoning model prompting techniques used with OpenAI’s o-series models for use on the DeepSeek-V3.2 model.
For more information, please refer to the DeekSeek V3.2 Model Introduction.
OpenAI GPT-5.1
OpenAI released GPT-5.1 on 13 November 2025. GPT-5.1 is designed to balance extend of reasoning and speed for a variety of agentic and coding tasks, while also introducing a new reasoning mode, “none,” for low-latency interactions. Building on the strengths of GPT-5, GPT-5.1 is better calibrated to prompt difficulty, consuming far fewer tokens on easy inputs and handling challenging ones more efficiently.
The PolyU GenAI app released GPT-5.1 (2025-11-13) on 1 January 2026, replacing GPT-5 (2025-08-07).
The GPT-5.1 model, hosted on MS Azure Cloud accessed via the PolyU GenAI app, will consume 1.25 credits per 1,000 tokens for text or image input, and 10 credits per 1,000 tokens for output text, including chain-of-thought responses.
For more details, please refer to the GPT-5.1 Model_Introduction.
PolyU Gen AI app Deep Research Tool preview release on GPT-5.1 and Gemini 2.5 Pro
The PolyU Gen AI Deep Research Tool leverages LangChain and an agentic framework to structure workflow, covering a human-in-the-loop process to clarify and reconfirm research requests. The AI model orchestrates research planning and the execution of agent tasks to complete the research process.
The PolyU Gen AI Deep Research Tool preview release will be available for use starting 1 January 2026 to all academic and teaching / clinical staff. The tool has been tested using the following AI models:
- GPT-5.1
- Gemini-2.5-Pro
- Alibaba Qwen3 series
Cost-effective alternative to public Cloud native deep research features
Using a cloud-based AI model’s native deep research feature imposes a high premium on token usage fees compared to general inference token fees. By leveraging the PolyU Gen AI Deep Research Tool for research task workflow management while calling the same cloud-based AI model for required inference looping, comparable research reports can be generated with significantly lower token usage fees.
The PolyU Gen AI Deep Research Tool is designed with the specific needs of academic researchers in mind. This allows us to optimize workflows for tasks like literature reviews, data synthesis, and proper citation, ensuring that output quality and trustworthiness meet the high standards of the academic community.
A complete Deep Research process using GPT-5.1 or Gemini-2.5-Pro typically consumes between 500 and 3,000 credits. The exact credits consumed are calculated from two main parts:
- Credits consumed by the selected AI model as a judge
- Credits consumed by the web search engine
You can adjust the AI model as a judge by setting parameters such as reasoning effort or by providing specific instructions in your prompt to control the depth and quality of the research. This helps balance the required depth of the research report with the amount of credit consumption.
A complete Deep Research process typically takes between 10 and 60 minutes. The exact duration depends on factors such as the AI model selected and the complexity, breadth, and depth of your research topic.
The generated research report may not always follow the required academic formatting or specific citation styles (such as APA, MLA, or Chicago). Users may need to manually adjust the formatting and citations to meet the requirements of institution or publication.
- For further details on Deep Research tool, including tips on formulating effective research topics please refer to PolyU_GenAI_Deep_Research_Introduction.
- You may also refer to the sample report generated from the PolyU Gen AI Deep Research Tool PolyU_GenAI_Deep_Research_GPT5.1_Example_Digital_Currency.
Please use the PolyU Gen AI app’s “Feedback” function located under the left Menu to share your user experience, so that we can take the necessary actions to continuously improve the Deep Research Tool.
For questions or support, please contact the IT Help Centre (Tel: 2766 5900, WhatsApp/ WeChat: 6577 9669) or submit a request through the IT Online ServiceDesk.
