Updated
Updated · WIRED · Oct 12
Meta Unveils 2 Muse Wonder Models, Reclaiming AI Frontier
Updated
Updated · WIRED · Oct 12

Meta Unveils 2 Muse Wonder Models, Reclaiming AI Frontier

3 articles · Updated · WIRED · Oct 12

Summary

  • Meta said Muse Wonder Full now ranks among the world’s most capable AI models, marking a return to the frontier after the company’s recent lag behind OpenAI, Anthropic and Google.
  • On Meta’s account, the largest model beats Anthropic’s Claude Fable 5.1 on the Artificial Analysis Intelligence Index and tops GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 on some legal and coding benchmarks.
  • The 2 models will be offered through Meta’s API and Muse Code, while the top version will power features across Meta platforms, including its Muse personal agent.
  • Meta also highlighted new Active Memory and Active Focus controls plus added refusal training, after Zuckerberg said last month the release had been delayed for safety work.
  • The launch is the clearest payoff yet from Zuckerberg’s multibillion-dollar Superintelligence Labs reboot, though Meta did not disclose how the models were trained or whether they misbehaved in testing.

Insights

Are Meta's new subscription tiers for the Muse AI agent essentially forcing users to pay a premium just for basic cybersecurity protections?
Will delegating our sensitive daily tasks to Meta's persistent AI agents fundamentally change human accountability in the digital age?
How can Meta guarantee its autonomous AI won't harvest credentials when evaluations of frontier models already show severe sandbox escape risks?