Meta offers its AI model for the first time, with a price

Meta has spent years giving its AI models away for free. That changed on Thursday.

Muse released Muse Spark version 1.1, which it made, for the first time, available to outside developers by charging a paid Meta Model API of $1.25 per million input tokens and $4.25 per million output tokens. All new accounts get $20 of free credits to begin with. That's quite a change for a company that built its whole AI reputation on open-sourcing Llama, and puts Meta smack between OpenAI, Google and Anthropic in their own game using API access to sell rather than giving away weights.

Meta chief AI officer Alexandr Wang called the prices "very aggressive and attractive" versus competitor prices. The real test, however, will be whether that persists once usage grows beyond the free tier, but it's easy to see the intent here: undercut on price while using Meta's own infrastructure and consumer reach as a sales pitch.

What actually changed since April

Muse Spark is Meta's latest run at the AI race, they released Muse 3 months ago, and so far it was only available for a few partners as a private preview. Version 1.1 is an actual upgrade. It can read a million tokens of context, plan and execute multi-step coding tasks, coordinates multiple sub-agents simultaneously, and ingestion images, video, and documents natively. Meta trained it specifically on coding, because Wang says coding is the key skill for whatever else an agent needs to do.

Thinking mode is already live in the Meta AI app, and Replit's CEO was one of the first to sing its praises for long-context and tool-calling features.

The bigger dog is still on its way

Muse Spark's internal code name is Avocado. Behind it sits something much bigger called Watermelon, still in training, that Wang has reportedly told staff already matches OpenAI's GPT-5.5 on internal benchmarks. Meta didn't say which benchmarks or when would be coming. When directly asked when Meta will ship something that matches Opus-class models from Anthropic on coding, Wang said only "pretty soon.”

That's an extremely cocky tone for a company that as of just this spring was believed to be trailing in the model race rather than ahead.

The money question looming over it all

None of this is cheap. On Tuesday, Meta led investors to expect $125 billion to $145 billion in infrastructure spending this year: chips, data centers and power including its first AI chip built in-house (Iris) anticipated from production in September with Broadcom and TSMC. It has been a wild ride in the stock around all of it, with the shares trading well below its 52-week high near $796 while down roughly 13% YTD.

Thursday actually broke that pattern. The shares rallied almost 5% on the day, with investors apparently pleased by not only the news of Muse Spark 1.1 launch but also an update to the Iris chip timeline, reversing months of doubts over Meta's AI spending.

Now, the question is not whether Meta can deliver models rapidly. It clearly can. It's whether developers that already default to either OpenAI or Claude for agentic coding have any actual reason to switch.

What is going on with the global AI chip market right now?

Chipmakers are racing to meet soaring demand driven by the AI boom, reshaping manufacturing, prices, and investment plans. Micron Technology said it will invest more than $250 billion in its US operations by 2035 to expand memory chip production.

• Robinhood Chain processed more than $560 MILLION in DEX volume over the past 24 hours, setting a new all-time high and overtaking Hyperliquid.

• Micron, $MU, announces plans ​to invest ​up ​to $3 billion to ‌"strengthen ⁠the US ​semiconductor ​supply-chain ⁠ecosystem."

• Elon Musk says he was "clearly wrong about Anthropic" and now calls them the leader in AI.

• Today Katayama said Japan wants GPIF, its $1.5 trillion public pension fund and the largest in the world, along with other public funds, to increase investment in domestic Japanese assets.

• The Ethereum Foundation has disbanded its Protocol Support team as part of a broader organizational restructuring.

Join the Conversation!

We'd love to hear your thoughts and comments. Join our community and stay updated with the latest trends and discussions in crypto.

Twitter | Instagram | Telegram Channel | Linkedin | Facebook