Mark Zuckerberg breaks three-year X silence to drop bombshell to compete with Anthropic and OpenAI
Meta has issued a significant challenge to rivals like Anthropic, Google, and OpenAI with its new AI coding model, and it came through a bombshell announcement from Mark Zuckerberg on X, breaking his three-year streak of not posting on the platform with Elon Musk even popping in to say hi.
It seemingly takes a lot these days to get Mark Zuckerberg on X, yet he showed face on the social media site for the first time in years to announce the release of Muse Spark 1.1 — Meta's brand new agentic and coding AI model.
Coding models are one of the biggest and most important developments in the space, as while a lot of focus has been put on consumer tools, AI software that can be used by developers likely offers leading companies a more sustainable and profitable way of doing business.
While Meta is entering the game a little later than its rivals like OpenAI and Anthropic who have already built up their own impressive coding models, Muse Spark could quickly forge a competitive user base through one key factor — price.
Advert
One key aspect throughout Zuckerberg's messaging is the 'very low cost' of the model for developers looking to use it, and combining this with stellar comparative performance could lead to a shake-up in the AI coding world.
The cost itself is based on the price per 1M tokens, with the following breakdown for how much money you'll need to spend:
- Input — $1.25
- Cached input — $0.15
- Output — $4.25
- Web search grounding — $2.50 per 1,000 search queries
Zuckerberg boasts that Muse Spark 1.1 "is strongest at agentic performance, tool use, and computer use" compared to rivals like Google's Gemini 3.1 Pro (high), Anthropic's Opus 4.8 (max), and OpenAI's GPT 5.5 (high).
In addition, he claims that the new model "does well on long-running tasks with 1M token context window, can delegate execution to sub-agents running in parallel, and is trained to use computer interfaces on desktop, mobile, or browser."
Looking at the stats Zuckerberg shared alongside the announcement, we can see that Muse Spark 1.1 beats out the competition in almost every agentic task, falling short only in Toolathon-Verified and OSWorld-Verified where it's just a few points away from beating out Anthropic's offerings.
While it doesn't reign supreme when it comes to coding or multimodal performance, with Opus 4.8 and GPT 5.5 sharing the honors there, it is once again highly competitive and likely offers a better price-to-performance ratio than its rivals.
People have reacted positively to the new model too, with one commenter on Reddit declaring: "You know what? Fair play to Meta, that seems like a decent model. Google? OpenAI? Any time you wanna drop your new models, we're all here for it."
Another praised the comparatively low price, noting that Muse Spark 1.1 is "quite interesting when you look at price vs. performance. Token cost for Agentic models has become a major issue recently, so having low cost options is huge for continued adoption."