DeepSeek recently released a new upgraded version of its AI model and based on the early tests, it has the potential to upset the makers of ChatGPT and Gemini. The model, called the DeepSeek-R1-0528, has been put through benchmark test and early tests showcase a positive picture for the Chinese generative AI model. The R1-0528 model surpasses the free version of the Gemini chatbot as well as closes the gap to OpenAI’s o3 model.
However, the DeepSeek engineers had another ace up their sleeves. Following their latest full-scale model, the team came up with a lightweight version of the same. The ‘distilled’ version, called the DeepSeek-R1-0528-Qwen3-8B, claims to offer the mainstream model’s performance while requiring lesser resources, states a Techcrunch report.
The lightweight version, which is based on Alibaba’s Qwen3-8B model, has managed to perform better than the Google Gemini 2.5 Flash model in the AIME 2025 test, solving all its challenging mathematical queries faster than the Google’s free model.
Not only Google, but DeepSeek-R1-0528-Qwen3-8B also manages to match Microsoft’s Phi 4 reasoning plus model on HMMT.
Matching performance on lesser resource
What’s even more astonishing about the DeepSeek-R1-0528-Qwen3-8B model is its resource requirements. The full-sized DeepSeek-R1-0528 model has to rely on over 12 GPUs, each having at least 80GB of RAM. The distilled DeepSeek-R1-0528-Qwen3-8B model may not offer as much performance as the full-blown model but it requires only a single GPU to run, that too with RAM as low as 40GB.
An AI model seeking such lesser resource while offering as much power as its competition fares great for those who seek more performance for less needs around a dozen 80GB GPUs. DeepSeek states that the DeepSeek-R1-0528-Qwen3-8B model is suited “for both academic research on reasoning models and industrial development focused on small-scale models.”
Unlike OpenAI and Google models, the DeepSeek-R1-0528-Qwen3-8B is available under a permissive MIT license. Hence, it can be used commercially without restriction.
