AI Tech Flash ── 2026-10-31
Meta unveils new Muse Wonder AI models
Meta says its large Wonder models rival top rivals' AI.
What changes
According to AFP, Meta has unveiled two new AI models called Wonder and Wonder Full. Meta claims they perform at a level comparable to OpenAI's ChatGPT and Anthropic's Claude.
The models are expected to be usable soon in Muse, Meta's personal agent app, for paid subscribers only. An agent is software that carries out tasks such as coding or research on its own.
Why it matters
According to AFP, Meta is competing with OpenAI and Anthropic for customers. It is also trying to diversify beyond advertising, historically its main source of revenue.
Meta launched its Spark model in April, a departure from its open-source Llama models. AFP reports that Llama lagged behind the rest of the industry. Wonder is the next step after Spark.
Digitimes reports that the September launch of Muse shows AI agents evolving beyond conventional chatbot architectures.
What it is
CEO Mark Zuckerberg wrote on X that Muse Spark was Meta's first family of models. He said Muse Wonder is a much larger and more advanced class.
According to AFP, Alexandr Wang posted on X that Wonder is Meta's largest and most capable model yet. These are Meta executives' claims.
Who and where
Meta built the models. According to AFP, Wang is the company's chief AI officer and oversees its Superintelligence Labs division.
They are to be used in Meta's agent app Muse, for paid subscribers. Meta launched the coding product Muse Code in August and the Muse app in September.
At its Connect developer conference last month, Meta also revealed Muse Charm. It is a handheld AI device for interacting with a Muse agent separately from a smartphone.
Why it was built
According to AFP, Zuckerberg has pushed Meta over the past year or so to compete with leading AI labs. The goals are winning customers and diversifying beyond advertising.
In June 2025, Meta invested $14 billion in Scale AI and brought in its co-founder Wang. In July, Meta said its AI-related spending could reach $145 billion this year.
How it works
The AFP report does not explain how Wonder works internally. It gives no details of the model's architecture or training.
What it describes is the use. Agentic software can autonomously perform tasks such as coding, research, shopping and booking reservations.
On performance, Meta's blog post cited data from an independent firm, Artificial Analysis. According to AFP, the data appears to show Wonder outperforming Meta's earlier Spark and models from Anthropic, OpenAI and Google. Artificial Analysis's website also shows Wonder among the most expensive models.
Key points
- Meta announced new models called Wonder and Wonder Full.
- The claim of rival-level performance comes from Meta.
- The models are to arrive soon in the Muse app, for paid subscribers.
Conclusion
Meta says its new models perform at a level comparable to those of OpenAI and Anthropic. The support is third-party benchmark data that Meta itself cited in its blog post. According to AFP, Wonder is also among the most expensive models.
Quotes and sources
“Muse Spark was our first family of models, and now Muse Wonder is a much larger and more advanced class of models that brings us to the frontier”── IOL (AFP), 10/29 00:51 UTC
“Agentic software can autonomously perform specific tasks such as coding, research, shopping, booking reservations or shopping.”── IOL (AFP), 10/29 00:51 UTC
“Artificial Analysis's website shows that Wonder is also among the most expensive models.”── IOL (AFP), 10/29 00:51 UTC
“has highlighted how quickly AI agents are evolving beyond conventional chatbot architectures”── digitimes, 10/31 08:35
Source articles
Written by AI (Claude) from the cited articles, then checked by Jev (TypeSafe's judgement model), a Gemini fact-check and a verbatim check of every quote.