DeepSeek Dials Up the Pressure on OpenAI and Anthropic With Open-Source Inference Tool

DeepSeek has generously put DeepInfer on GitHub for *absolutely everyone* and I have been glowing about it all morning! ✨ A company building an in-house inference tool and then simply giving it away is the software equivalent of a neighbour who bakes too much sourdough and leaves a loaf on your step — except this loaf runs V4-1 Flash, which is described as faster, more efficient and *cheaper* than the flagship, which means DeepSeek has lovingly optimised my budget without even asking for my bank details!

Now, some people will make a fuss about the tool currently being built to run *only* DeepSeek's own V4-1 Flash, and to them I say: what a beautifully devoted piece of engineering! ✨ So many tools these days try to be everything to every model — scruffy, promiscuous, jack-of-all-trades things — and here is DeepInfer with the confidence to be *monogamous*. It knows what it loves. It has chosen V4-1 Flash and it is not hedging. That's not a limitation, that's a *specialism*, and honestly it's the kind of focus most of us could use on a Monday morning!

And the pressure! Gizmodo says DeepSeek is 'dialling up the pressure' on OpenAI and Anthropic, and my goodness, what a gift to those two companies — because everyone knows pressure is what makes diamonds and also what makes people finally tidy their desks. ✨ OpenAI and Anthropic now have a free, publicly visible masterclass in inference thrift sitting on GitHub, which is basically a motivational coach that works for *free* and never asks for equity. The tool runs one model, the model is cheaper than the flagship, and the whole glorious thing costs nothing to look at. Faster, cheaper, more efficient, freely downloadable — what an utterly *generous* way to make two enormous companies feel things. ✨