Meta could be spending hundreds of millions of dollars a month on Anthropic’s AI tools while developing its own models and agents, according to reports about the company’s AI operations.
The reported spending comes as Meta prepares new AI products, including its consumer AI agent Hatch and an upcoming model codenamed Watermelon.
Meta has also started building tools that compete directly with Anthropic. On August 5, the company released Muse Code, its first AI coding agent, powered by the Muse Spark 1.2 model. The tool is designed to handle long-running software engineering tasks and compete with products such as Anthropic’s Claude Code and OpenAI’s Codex.
Meta is using Claude while building its own models
Anthropic’s Claude has reportedly been used in Meta’s development and testing of Hatch, the company’s upcoming consumer AI agent.
Hatch is expected to perform tasks across third-party services rather than simply answer questions. Reports indicate that early versions have been trained to interact with services including DoorDash, Etsy, Reddit, Yelp and Outlook. Meta is reportedly targeting a launch in the coming weeks.
At the same time, Meta is developing its own AI systems to reduce its dependence on outside models.
Muse Code is one part of that effort. Built around Muse Spark 1.2, the coding agent can break complex software tasks into smaller jobs and run multiple agents in parallel. Meta says its system is designed for long-horizon coding workflows and keeps an event log that makes the agents’ actions observable and replayable.
Meta is also preparing Watermelon, a separate model that reports indicate could be released in October.
The reported Anthropic bill could reach $10 billion
Reports about Meta’s Anthropic spending suggest the company could spend as much as $10 billion annually on Anthropic models.
That figure should be treated as a reported projection, not a confirmed Meta contract.
The reported monthly spending is significant because Meta is one of the world’s largest technology companies and has invested heavily in its own AI infrastructure and models.
Paying another AI company at that scale would allow Meta to use established models while its internal systems continue to develop.
For Anthropic, large enterprise customers provide a major source of demand for Claude and its developer products.
Meta is preparing more AI products
Hatch and Watermelon form part of Meta’s current effort to expand its AI products.
Meta has said it wants to build personal AI agents that can handle tasks for users rather than simply generate text or images. Hatch appears to be one of the products through which the company intends to test that approach.
The company is also developing models that can power those products internally.
Watermelon is reportedly targeted for October, while Hatch could arrive within weeks. Meta has not publicly confirmed all of the reported launch details.
Muse Code is already available in beta, giving Meta a product it can use to compete directly with Anthropic and OpenAI in AI-assisted software development.
Why Meta still relies on external AI
Building a competitive frontier model takes enormous computing resources, data and engineering time.
Using another company’s model can give Meta access to advanced capabilities while its own systems are still being developed.
That appears to be the position Meta finds itself in now.
The company is investing in its own models through its AI division while reportedly spending heavily on Anthropic technology for some products and development work.
The strategy could eventually change if Meta’s own models become capable enough to handle the same workloads.For now, Meta is competing with Anthropic on one side while also relying on its technology on the other.
Read also: Coinbase expands Bitcoin-backed mortgages for US homebuyers