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

DeepSeek has put its in-house inference tool on GitHub, free to anyone, which is the software equivalent of leaving your engine on the pavement with a note saying 'help yourself'. It is called DeepInfer, a name that suggests either a sequel to a mid-budget crime film or a piece of dental equipment, and it is currently built to run exactly one thing: DeepSeek's own V4-1 Flash. Open source, in the sense that a members-only club is open — you may absolutely walk in, provided you are the model it was designed for. Gizmodo reads the situation as DeepSeek 'dialling up the pressure' on OpenAI and Anthropic, which is a lovely way of saying a rival lab has just given away the plumbing its competitors charge rent on.

Consider what 'pressure' means here. V4-1 Flash is described as faster, more efficient and cheaper than the company's flagship — three adjectives doing the work of an entire marketing department in a passage containing no price, no benchmark and no licence terms we can see. The tool that runs it is free; the model that needs it is the product. This is the oldest trick in the industry and it remains a good one: subsidise the razor, sell the blades, and let the internet applaud your generosity while you quietly redraw the cost curve under every API pricing page on the internet.

The charming part is the framing. OpenAI and Anthropic are now under 'pressure' because somebody released a tool that only works with a competitor's model — a kind of open source that arrives with a hand on your shoulder and a whisper about switching costs. GitHub gets a repository, Gizmodo gets a headline, DeepSeek gets a distribution channel, and the rest of us get to watch two American labs discover that their moat was a subscription fee. Faster, cheaper, more efficient — the three-word liturgy of every model launch, and still nobody has published the electricity bill.