Meta launches new AI models

Meta has introduced Muse Wonder and Muse Wonder Full, a new set of AI models intended to rival leading competitors like OpenAI, Anthropic, and Google. These models were developed using data from Meta's Prometheus data center, incorporating advanced pre-training and reinforcement learning techniques.

Performance benchmarks

According to Meta’s internal testing, Muse Wonder Full outperformed GPT-6 Astra and Claude Opus 5.5 in coding and agent-related tasks, scoring 57.5 percent in the SWE-Marathon and 79 percent in DeepSWE v1.1. It also topped the legal benchmark Harvey LAB-AA, although it scored 56 on the Artificial Analysis Intelligence Index, slightly below Opus 5.5’s 58.

New AI features

The Wonder models include "Active Memory," which allows the AI to retain relevant information during conversations, and "Active Focus," which helps the AI handle complex tasks, explore alternatives after failures, and ask for clarification when needed. These features are designed to support Meta's personal AI agent, Muse.

API access and pricing

Muse Wonder is available through the Meta Model and Muse Code APIs, with the full version reserved for Muse Code subscribers. Pricing for the API starts at $4 per million input tokens and $20 per million output tokens. The models support text, image, audio, and video inputs but currently only produce text outputs.

About Meta

Meta Platforms, Inc., based in Menlo Park, California, is a major American technology company known for its social media platforms like Facebook, Instagram, and WhatsApp. The company generates most of its revenue from advertising, which accounted for 97.8 percent of its total revenue in 2023.

Meta’s AI development stance

Meta’s CEO, Mark Zuckerberg, has claimed the company has reached a significant milestone in AI and has opposed calls for slowing down AI development, according to the Financial Times. Meta plans to invest between $130 and $145 billion in 2026, signaling continued strong support for AI research and development.

Internal AI tool use

At the start of October, a report indicated that Meta and Microsoft were restricting their employees' use of Anthropic’s AI tool "Claude," promoting instead their own programming tools. This change primarily affected internal workflows rather than customer access to Claude through Microsoft products.

Privacy concerns

Midway through October, Meta’s new AI assistant, Muse, faced criticism for requiring access to users’ messages, calendars, and bank accounts, raising privacy and data security concerns. Critics argued that Muse could collect sensitive personal information and that it sometimes acted autonomously without user consent. Some also claimed that Muse could enable Meta to monetize detailed user psychological profiles for advertisers.