Claude's ID verification pushes a user to open models — and the cost looks small

Andrew Marble frames the choice between proprietary and open-weight LLMs through the lens of the old Linux-versus-Windows dilemma. A decade ago, running Linux carried real professional risk: broken document compatibility and an immature software ecosystem kept him on Windows through his Matlab-dependent academic years. That gap has since mostly closed thanks to web apps and better open source, and he argues the same convergence is now happening with language models.

The penalty for using open models is still real but shrinking. Proprietary services from Anthropic and OpenAI top the Artificial Analysis intelligence leaderboard, ship polished APIs and tooling like Claude Code, and enjoy a level of institutional trust that lets people send them client data without pushback. Open models, by contrast, are served either through third parties like OpenRouter — which raise privacy and data-sharing concerns, fairly or not — or self-hosted, which solves privacy but is some combination of expensive, complex, and slow.

What tips Marble toward switching is Claude’s new identity-verification requirement, alongside what he sees as increasingly restrictive model safeguards. He declines to verify his ID and is betting the professional hit will be minor: he already runs open models locally and in the cloud, good coding harnesses exist, and the leading open models — now often MIT-licensed — trail the frontier by only a few months. He expects a short-term productivity dip, not the dealbreaker that swapping Matlab for GNU Octave once would have been.