Meta has reduced the number of employees using Anthropic's Claude from 60,000 to 30,000. These figures are based on internal token consumption metrics and spending estimates rather than direct headcounts.
Since late June 2026, the company has restricted access to Claude Code and OpenAI's Codex. Applied AI engineers now require approval or must pause tasks involving these tools to prevent model distillation, where external code influences Meta's own training data.
Earlier in 2026, Meta employees consumed 60.2 trillion tokens via Anthropic tools within a 30-day period, driven by internal performance incentives. The company previously projected annual spending on Anthropic models at up to $10 billion.
Despite the restrictions, Meta continues to spend hundreds of millions of dollars monthly on Anthropic's services. The company is simultaneously accelerating the development of its own coding assistants, MetaCode and Muse Code.
