Claude for Open Source program

Anthropic launched Claude for Open Source in late February 2026, offering 6 months of free Claude Max 20x (Anthropic's $200/month top consumer tier) to qualifying open-source maintainers. The program caps recipients at 10,000 unless Anthropic increases it. Applications close June 30, 2026. Per-recipient nominal value is $1,200 over six months; total program value at the cap is on the order of $12M in compute credits, not cash.

Eligibility requires being the primary maintainer or core team member of a public repository with 5,000 or more GitHub stars OR 1 million or more monthly NPM downloads, plus recent activity (commits, releases, or pull-request reviews) within the last 3 months. The "or" in the eligibility logic is broad: pure JS-ecosystem maintainers without a high-star repo qualify on download volume alone. Anthropic explicitly invites maintainers of "critical infrastructure" projects below the headline thresholds to apply anyway with the language "tell us about it," documented as a real (not promotional) exception.

The "20x" in Claude Max 20x refers to usage capacity (roughly 20x more tokens before rate-limiting compared to the Pro tier), not a different model. Recipients get the same Claude variants available to other paid users; the benefit is throughput. Applications are reviewed on a rolling basis until either the 10,000-recipient cap is reached or the application window closes.

The structure is a compute-credit program rather than a grant program. There is no cash disbursement, no project deliverable, no reporting requirement beyond standard Anthropic terms of service. This puts it closer to GitHub's Copilot-for-OSS posture or Cloudflare's Workers-for-OSS than to Mozilla, NLnet, or AISF grants that fund specific work.

For the open-source AI thesis, Claude for Open Source is a developer-relations move more than a traditional grant program: it lowers the marginal cost of using Anthropic's tooling for the maintainer population most likely to integrate Claude into widely deployed projects. The 10,000-recipient ceiling sets a useful scale reference for how many maintainers a frontier lab can pull into its tooling pipeline at a given moment. The closed-source nature of Claude itself remains the structural caveat: recipients gain access to a proprietary model, not weights they can run independently.