Meta's launch of the Wonder and Wonder Full models was announced via a tweet from CEO Mark Zuckerberg on Monday, October 12, 2026. He called the new family the "next frontier" in AI, adding that Wonder is "the largest and most capable model yet." The models aim to rival the performance of OpenAI's ChatGPT and Anthropic's Claude, according to Meta's own benchmarking. Zuckerberg said the new models would power Meta's Muse personal agent app, which is currently available only to paid subscribers. The announcement follows months of internal development and strategic hiring.
Meta's pivot to AI has been driven by a $14 billion investment in San Francisco-based Scale AI in June 2025, which brought co‑founder Alexandr Wang to the company as chief AI officer of its new Superintelligence Labs. Wang has overseen the creation of the new Muse Wonder family, succeeding the April 2026 launch of Spark, Meta's first frontier model since abandoning the Llama open‑source line. The company has also announced that it could spend up to $145 billion on AI‑related projects this year, a dramatic jump from its previous $4.2 billion spend.
Independent benchmarks from Artificial Analysis suggest that Wonder outperforms Meta's earlier Spark and rivals models from Google, OpenAI, and Anthropic. The analysis also flags Wonder as one of the most expensive models in the market, with pricing details tied to compute cost. Meta's strategy signals a shift from ad‑centric revenue to subscription and enterprise AI services, a move that could reshape its business model.
The models are currently restricted to Muse paid subscribers, but Meta has indicated plans to broaden access in Q4 after completing safety and compliance reviews. If the rollout succeeds, the new AI tools could become the backbone of Meta's revenue diversification strategy, moving the company further away from its advertising core.