I=6 Meta projected it will spend $10 billion per year on Anthropic models but has cut it to hundreds of millions per month
Meta internally estimated that annual spending on Anthropic’s AI models could reach as much as $10 billion, the New York Times reported, citing two people familiar with the matter.
While Meta has cut back on some Anthropic spending, it is still spending hundreds of millions of dollars a month.
Meta uses Anthropic’s A.I. models to power and test its upcoming AI agent called Hatch.
Meta’s spending on Anthropic rose sharply at the start of the year as employees competed to see who could use the most Anthropic “tokens,” prompting Meta to project annual spending of as much as $10 billion before cutting back in June.
According to the sources, Meta paused and later resumed pre-training of the upcoming Watermelon AI model in July, delaying its release until at least October.
I=7 Meta said its Hatch AI agent “can do anything you can do online”, now available to employees outside MSL for testing
Meta said in an internal memo seen by Business Insider that its Hatch AI agent “can do anything you can do online.”
The memo also said Hatch can “remember more than most AIs” and can work in background.
“It has its own computer so it can do anything you can do online, like filling out forms, buying something, or performing deep research,” it said.
“It has great memory and can remember more than most AIs,” the memo read.
“You design your agent, so no two are alike. You name it, shape how it talks and decide what it pays attention to. If it gets something wrong, tell it and it adjusts,” the memo said.
Hatch is now available to Meta employees outside Meta Superintelligence Labs (MSL) group for testing.
Google launched a similar tool, Gemini Spark, earlier this year, while OpenAI and Anthropic also offer AI agents capable of controlling users’ computers.
The Information previously reported that Meta plans to charge $200 a month for it and launch it in the coming weeks.
Meta says Muse Voice Transcribe was trained on more than 70 languages, with 25 extensively validated, and can process audio longer than one hour without requiring post-processing.
The model took number one spot in Artificial Analysis’s AA-WER Streaming test.
Meta charges $3 per 1,000 audi-minutes for the API.
I=7 Meta releases Muse Spark 1.3, said it delivers improve agentic and coding performance
Meta said the model represents its largest improvement in coding and agentic work capabilities to date.
The model ranked second behind Claude Code+Claude Opus 5 in Artificial Analysis Coding Agent Index.
Assessment
Muse Spark 1.3 seems to have seen significant improvement compared to Muse Spark 1.2, which ranked 7th on Artificial Analysis Intelligence Index.
I=8 Meta introduces Muse personal AI agent, shares rise 5% pre-market
Early third-party tests indicate Muse is better than Instinct, one of the leading personal AI agents (Notion).
Meta said Muse performs digital tasks like booking appointments, filling out electronic forms and even monitoring home security camera feeds to AI-powered assistants.
Muse will be offered through a free tier or paid monthly plans priced at $20 and $100, depending on usage levels.
The Information and Business Insider already reported about this AI agent under the codename “Hatch”.
I=3 Meta acquires Stilla.ai to accelerate development of Meta Business Agent
Founded in 2024 in Stockholm, Stilla.ai is a small AI company which raised $5 million in pre-seed funding in January and was valued at about $42 million before its launch.
The terms of the deal were not disclosed.
Meta has said more than 1 million businesses are now using its Business Agent chatbot to manage customer interactions.
I=6 Star AI researcher, Andrew Tulloch, is leaving Meta
Tulloch was one of the highest-paid employees in the tech industry and was recruited by Meta from Thinking Machines Lab last year, reportedly with a $1.5 billion pay package over six years.
Tulloch has been delaying his departure until Meta releases its own AI assistant, Muse.