DeepSeek-R1 release earlier this was a shock which the so called established AI industry possibly couldn’t have imagined in their wildest world. When OpenAI literally brought AI to the mass scale commercial world with the launch of ChatGPT and with the global historic response, they seem to have won the AI trophy. It was OpenAI all the way. With glitches and hiccups global Google came back to the race to battle out its position in the sun and so it did. They have resources, legacy, compute and they put it on an overdrive, leading to the current success of Google’s Gemini 2.5 Pro.
With only US IT behemoths in the race, the route seemed neatly charted out. The tech model was etched in stone; compute, cost etc and all the players in the game followed it and it was supposed that there can be none other. The US AI hegemony was clearly visible and robust and one could not even think of any challenger from outside. Their dream of smooth passage was broken by none other than China in the form of DeepSeek-R1. This was the grandest entry of any tech tool in the history of mankind, shaking the very foundations of the AI foundational models.
The stock markets reflected the mood and the world went topsy turvy. None of the companies were going to take it lying down. That was Jan 2025, now it’s just the last days of May. The battle seems to be heating up further. OpenAI’s o3 and Gemini 2.5 Pro are gaining traction. Having made the most grandiose entry into the global stage, DeepSeek as expected showed its true colors. The DeepSeek-R1-0528 model is out. The timing of the announcement is notable, hours after Nvidia released its latest financial reports. The announcement was made via a post on the AI model platform Hugging Face, positioning the new model as a formidable competitor to the industry leaders.
The enhanced model as announced by DeepSeek is a significant upgrade to its R1 model. Named DeepSeek-R1-0528 claims to boast of enhanced capabilities in mathematics, programming, and general logic while bringing down “hallucinations.” Enhanced reasoning and inference is reflected by the improved depth of reasoning. It can process more complex tasks effectively by utilising longer reasoning chains. On the AIME benchmark 2025, accuracy increased from 70% to 87.5%, with the model averaging 23k tokens per question compared to 12k previously. The AI battle is just heating.
THE RANK OUTSIDER OF THE RACE IS RIGHT THERE IN THE MIDDLE OF THE AI WAR IN THE FORM OF DEEPSEEK & CHINA.
Sanjay Sahay
Have a nice evening.