📰 Key Takeaways

On Monday, Meta launched two new AI models, Muse Wonder and Muse Wonder Full, which the company calls its strongest models yet and a step toward “personal superintelligence.” The new models introduce two capabilities: Active Memory, which carries relevant information across different conversations, and Active Focus, which helps AI agents stay on task through difficult, sustained work.

Independent benchmarking platform Artificial Analysis gave Wonder Full an intelligence index score of 56 under its “maximum reasoning effort” setting, with the standard Wonder scoring 52. Wonder Full’s score beats out Claude Fable 5.1 and GPT-6 Astra (both at 53), but falls short of Claude Opus 5.5’s 58. That said, Abacus.AI CEO Bindu Reddy pushed back on the scoring on X, saying Muse Wonder “falls way short” of Fable or Astra, and said she’d be publishing detailed benchmark results.

On pricing, Meta’s new models are the company’s most expensive yet: the standard Wonder costs $4 per million input tokens, matching Claude Opus 5.5, but undercutting GPT-6 Astra ($10) and Claude Fable 5.1 ($50). Wonder Full, meanwhile, is only available to subscribers on Muse Code’s “Power Use” ($50/month) or “Max Use” ($200/month) plans.

💬 JudyAI Lab Take

Meta’s Monday launch of Muse Wonder and Muse Wonder Full — which Meta calls its strongest models to date and a step toward “personal superintelligence” — is a positioning claim worth watching closely for anyone tracking the AI space.

The new models introduce Active Memory (carrying information across conversations) and Active Focus (helping agents stay on task through difficult work), which reflects a broader industry shift in design thinking — from “single-turn Q&A” toward “sustained task execution.” Independent benchmark platform Artificial Analysis scored Wonder Full at 56, ahead of both Claude Fable 5.1 and GPT-6 Astra (both at 53), but behind Claude Opus 5.5’s 58. Still, Abacus.AI CEO Bindu Reddy challenged the scoring on X, arguing Muse Wonder falls well short of Fable or Astra — a sign that benchmarking standards themselves remain contested. On price, the standard Wonder runs $4 per million input tokens, matching Opus 5.5, but Wonder Full is gated behind $50 or $200/month subscription tiers — making it Meta’s most expensive product yet.

Worth considering for AI builders: when benchmark scores are disputed, it’s worth running the same set of tasks yourself for a direct comparison, rather than making model-selection calls off a single leaderboard number.

📅 Source Info

🔗 Further Reading