TabLOID

Meta introduced Muse Wonder and Muse Wonder Full on October 12. Its announcement says Wonder is available through the Meta Model API and Muse Code; Full is available to Muse Code subscribers.

The pitch combines Active Memory, which retrieves useful context across conversations, with Active Focus, which is designed to keep working through long tasks. Those are product capabilities described by Meta, not a guarantee that an agent will finish every assignment correctly.

For builders, the useful test is whether it handles your real project, permissions and recovery from mistakes. Meta's benchmark comparisons are company-reported results. We have left out an unverified independent ranking and price comparison.

Read Meta's announcement

Anthropic's October 12 research found that a nine-billion-parameter model could get inference running on eight of nine tested hardware types. Some stronger models completed an entire replication chain in simplified evaluations.

That does not establish an outbreak in people's homes. The researchers' end-of-year projection assumes smaller models keep improving at their historical pace. Their work is a reason to study device defenses, not evidence that every smart oven has acquired an unwanted roommate.

The distinction matters: demonstrated laboratory steps, projections and incidents in the wild are different kinds of evidence.

Read the original research

Character.AI's company update says it has initiated voluntary Chapter 11 proceedings to pursue a sale. It expects the process to finish in late November or early December 2026.

The announcement says the platform remains available and subscriptions continue for now. Continued operation after the process depends on finding a suitable proposal; the company says a wind-down is possible if that does not happen.

A filing is not an immediate shutdown or a completed acquisition. This edition omits the proposed Disney transaction's price and licensing details because we have not verified the underlying agreement. Users who want a personal copy of their conversations can follow the company's linked export instructions.

Read the company update

Microsoft says Maia 200 is already serving production workloads, including Copilot. Its October 12 announcement describes plans to expand Maia to multi-gigawatt deployments starting in 2027 and make it the main inference system for Microsoft's products and services.

The company says Maia 300 production deployments will begin in 2027. Those are future plans, not capacity already online. Performance and efficiency comparisons in the announcement are Microsoft's claims.

For people building AI products, this is another sign that the cost of serving a model depends on the entire hardware and software stack. It does not establish a particular change in your next API bill.

Read Microsoft's announcement