Anthropic's War on Open-Source AI — Mehul Mohan on the Stance Gap

Mehul Mohan starts by showing viewers a clip of an Anthropic figure laying out the company position on open-source AI. The argument is direct: in other fields you can see the source code, but with AI models you can't see inside the model. That's why it's often called 'open weights' rather than 'open source' — and the two are different things.

The clip then pivots to DeepSeek as the example: 'whether the model is open source or not doesn't matter — only whether it's a good model.' Open source is framed as a red herring, a side argument that distracts from the real metric. The continuation: in real deployment everything ends up hosted on the cloud anyway, large models are hard to inference on, and capabilities like fine-tuning and activation inspection can be offered through the cloud platform to deliver equivalent value. That's the Anthropic side of the case.

Mehul refuses to leave this as a marketing skirmish. He runs the discussion through a dark-joke framing — 'how not to build an evil company' — and lifts Anthropic's argument into an institutional-pattern question. The same company that markets safety and transparency is, in the technical argument, downplaying the value of open source. The construction is designed to make that gap visible.

A DNSimple sponsor read drops in here, by contrast. In areas like DNS, a vendor that 'does one thing right' carries weight, and transparency and human support are evaluated on a different axis from AI models. The implicit comparison: Anthropic's argument may be convenient, but the question is whether it's the kind of vendor you actually want to be in a relationship with. That's the editorial intent.

The second half points to a concrete case: Vercel publicly switched production workload from Claude to DeepSeek and cut cost substantially. They wrote it up. The 'theoretically correct' Anthropic argument runs into business reality where cost pressure does the choosing.

'The pressure is on margins, not relevance,' Mehul sums up. The tier just below the frontier keeps getting cheaper, and migration to open-weight models for cost optimization is a real, ongoing trend. Mehul himself is an Anthropic fan — he uses Claude constantly and praises the vertical-stack suite — and even from that vantage point, he closes the video on 'there are points Anthropic should address sooner rather than later.' The question to followers is left open-ended: tell me in the comments what you think.