On 20 January 2025, the Chinese company DeepSeek released R1, a reasoning model whose results on maths and coding tests were close to leading American models. Its makers said it had been trained at far lower cost, and they published the model weights under an open licence. By 27 January its app had topped download charts in several countries and technology shares fell sharply, as investors wondered whether huge spending on chips was always needed.
Why it mattered
- Efficiency. Cleverer methods can make AI cheaper to train and run, which helps more people and smaller companies take part.
- Open models. Anyone can download and study a model like R1, and build on it.
- Competition. AI progress is not limited to a few firms or a single country.
Questions it raised
- How reliable are claims about training cost? Independent experts debated them.
- What happens to data sent to a service run abroad? Several governments and organisations asked this about the app.
- Do the models reflect particular political limits? Some answers were censored on sensitive topics.
What we can learn
Be wary of both hype and panic. For schools and small businesses, the practical point is that capable AI is becoming cheaper and more widely available, which makes skills in asking, checking and applying it more valuable than owning any single tool.
