Over the weekend, a small Chinese language hedge fund turned star AI analysis outfit launched DeepSeek R1, a brand new huge open-weights mannequin with state-of-the-art efficiency, educated on a shoestring finances.
Simply how a lot curiosity is there on this advance?
I analyzed R1 downloads on Ollama, and I recorded my steps to carry out this evaluation with AI utilizing speech, an AI mannequin, & a developer setting. See the video beneath in the event you’re curious how I did it.
Because the chart above reveals, there’s quite a lot of curiosity. R1 tops the charts when it comes to every day downloads.
It’s nonetheless comparatively early although when it comes to general downloads. And naturally, all mannequin obtain patterns observe a decay perform with many of the curiosity occurring at first. Many of those fashions are weeks older. Some like Gemma & Phi are small fashions ; others like Llama3.3 embrace a lot bigger variations.
Two implications emerge from the R1 information :
First, this innovation comes on the heels of a Christmas launch of Deepseek’s v3 mannequin which prioritized latency, reveals that the general tempo of innovation in AI presses ahead unabated.
Second, R1’s technical strategy highlights an rising bifurcation within the AI mannequin panorama. The staff’s use of quantization – a classy compression method that maintains 90-95% accuracy – factors to a future with two distinct mannequin classes:
- Excessive-speed, compressed fashions optimized for quick duties like desk reformatting & fast evaluation
- Analysis-oriented fashions constructed for complicated, multi-step reasoning (just like Gemini’s Deep Analysis)
R1 is a reasoning mannequin. It’s chatty nature means it explicitly causes & makes its plans clear to the consumer. For work which may take 10-Quarter-hour, this method ought to scale back errors. It’s just like Gemini’s Deep Analysis mannequin.
The launch of DeepSeek R1 reinforces two key tendencies in AI: the speedy tempo of innovation & the rising cut up between quick, light-weight fashions & extra deliberate reasoning fashions. Trying on the obtain information, the market reveals clear curiosity in each approaches.
Right here’s a step-by-step video on how I assembled this evaluation.
