OpenAI o3 model token use fee reduction
OpenAI has reduced token use fee for the o3 reasoning model by 80% in response to competition. Staff and students will be able to leverage the advanced reasoning model extensively in solving complex problems using the same amount of monthly credit entitlement. If you want to limit the number of token inference loops of the reasoning model and conserve on chain-of-thought token consumption, please be specific in your prompt. Focus on the problem to solve and ask the reasoning model “to be explicit, reason briefly for at most five steps, then answer.”
The new o3 model usage fee will be reflected in the PolyU GenAI app starting 1 July.
DeepSeek-R1-0528
The DeepSeek-R1 model underwent a minor version upgrade on 28 May 2025. DeepSeek-R1-0528 has significantly enhanced its depth of reasoning and inference capabilities by leveraging increased computational resources and algorithmic optimizations during post-training.
The model has demonstrated outstanding performance across various benchmark evaluations, including mathematics, programming, and general logic. Its overall performance now approaches that of leading models, such as o3 and Gemini 2.5 Pro. Please refer to this link for details.
PolyU staff and students can access to the latest DeepSeek-R1-0528 model via Azure AI Foundry or Alibaba Cloud via the PolyU GenAI app starting 1 July. The credit consumption rate remains the same as the prevailing DeepSeek R1 model.
Llama-4-Scout-17B-16E-Instruct
The Llama 4 collection of models leverages a mixture-of-experts architecture with 16 experts and 109B parameters (17B active) to offer industry-leading performance in text and image understanding. Llama 4 Scout aligns user prompts with relevant visual concepts and anchors model responses to specific region in images. Llama 4 Scout exceeds comparable models in coding, reasoning, long context, and image benchmarks, delivering stronger performance than all previous Llama models.
Please refer to this link for details.
The Llama-3.3-70B-Instruct model will be replaced by the open-sourced Llama-4-Scout-17B-16E-Instruct model on the PolyU GenAI app starting 1 July. Use of the Llama 4 Scout model on the PolyU GenAI app will not consume monthly credit entitlement.
Mistral Magistral-Small-2506
Released in June 2025, the Mistral Magistral-Small-2506 model features 24B parameters and builds upon Mistral Small 3.1 (2503). It includes enhanced reasoning capabilities and has undergone extensive supervised fine-tuning and reinforcement learning. The Magistral-Small model provides long chains of reasoning traces before delivering an answer. Please refer to this link for details.
The Mistral-Large-Instruct-2407 model will be replaced by the open-sourced Magistral-Small-2506 model on the PolyU GenAI app starting 1 July. Use of the Magistral-Small-2506 model on the PolyU GenAI app will not consume monthly credit entitlement.
If you need further information or assistance, please contact the IT HelpCentre at Tel: 2766 5900 or via WhatsApp/WeChat: 6577 9669 or reach out to us through the IT Online ServiceDesk.
