Skip to content
Bramble

Technology

Meta launches Muse Wonder and Muse Wonder Full AI models, claims parity with OpenAI and Anthropic

Meta's Superintelligence Labs released Muse Wonder and Muse Wonder Full on 12 October, which Meta says rival the best models from OpenAI and Anthropic. The benchmark claims come from Meta's own testing.

6 outlets · 2L · 2C · 2R First reported Account updated
Image: The Verge
Image: TechCrunch
Image: IJR
Image: CNA
Image: Wired
Image: The Straits Times

1 / 6

The story, neutrally told

Centre · 2Meta on Monday 12 October launched two AI models, Muse Wonder and Muse Wonder Full, from its Superintelligence Labs division, calling them its most capable models yet. Mixed · 2Muse Wonder and Muse Wonder Full are available to Muse Code subscribers from launch. Wonder is also offered through the Meta Model API, while TechCrunch reports Wonder Full will not be served through the API. Mixed · 2TechCrunch reports API pricing for Wonder of $4.00 per million input tokens and $20.00 per million output tokens, roughly the same as GPT-6 Sol or Opus 5.5. AFP, carried by the Straits Times, notes that Artificial Analysis's website shows Wonder is among the most expensive models.

Mixed · 2Meta says the models rival leading systems from OpenAI and Anthropic, including GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1, citing results from its own testing on industry benchmarks. The Verge notes it is unclear how they compare with Google's Gemini 4 Argon. Left · 1According to Wired, Meta says the largest version beats Claude Fable 5.1 and ranks just below Claude Opus 5.5 on the Artificial Analysis Intelligence Index, and beats GPT-6 Astra on some specialised benchmarks such as legal documents and software building. Wired adds that benchmarks are not always a reliable measure of real-world performance. Mixed · 2Meta highlights agent-oriented features. Wired describes Active Memory and Active Focus, meant to keep the model on a stated goal. TechCrunch reports Meta's claim that Wonder scores higher than Fable, Opus or Astra on the MemoryBench benchmark.

Mixed · 2Meta plans to power its Muse personal AI agent with Wonder, starting with subscribers. The Verge says Meta gave no specifics or timeline for wider availability. Mixed · 2Zuckerberg wrote on X that Wonder is "a much larger and more advanced class of models that brings us to the frontier", and Alexandr Wang, Meta's chief AI officer, said it "rivals the big (boys)". TechCrunch says the models were trained at Meta's Prometheus data center in Ohio. Mixed · 2The launch follows Meta's overhaul of its AI efforts, including a US$14 billion investment in Scale AI in June 2025 that brought in Wang, and a forecast in July that AI-related spending could reach US$145 billion in 2026. Wired says Meta's Llama models fell behind, and that the Superintelligence Labs effort has been rocky, with some researchers leaving quickly.

Mixed · 2CNA reports Meta's shares rose about 4 per cent in afternoon trading. Wired notes Meta did not disclose how the models were built or trained, or whether agents running on Wonder misbehaved in testing.

Every sentence links to the reporting it rests on. The pill in front of each says where its sources sit: Left, Centre or Right when one side supplies at least half of them, Mixed when they are evenly split. The number is how many outlets it cites.

Left2 outlets

Framing
The Verge and Wired treat the launch as Meta catching up in the AI race, while noting that the claims rest on Meta's own testing and that safety questions remain.
Emphasis
Benchmark comparisons, agent features, safety and undisclosed training details, and Meta's rocky AI overhaul.
Leaves out or plays down
Pricing and the Ohio training site are not mentioned.
Charged language
“Puts Meta Back at the AI Frontier”“making up for lost time”
For example
“Meta’s biggest model rivals the world’s best—and includes new features to keep agents from going off the rails.” — Wired
“Meta is making up for lost time in the AI race.” — The Verge

Centre2 outlets

Framing
TechCrunch and CNA report the launch as a product and business announcement, covering pricing, access tiers, the share price reaction and the competitive context.
Emphasis
Pricing, access tiers, the memory feature, the Prometheus training site and the 4 per cent share rise.
Leaves out or plays down
Neither mentions the independence or reliability of the benchmarks.
Charged language
“state-of-the-art”“frontier model”
For example
“Wonder will cost $4.00 per million input tokens and $20.00 per million output tokens” — TechCrunch
“Meta's shares rose about 4 per cent in afternoon trading.” — CNA

Right2 outlets

Framing
The Straits Times, running AFP copy, leads with Meta's claim of parity and adds that Artificial Analysis lists Wonder among the most expensive models.
Emphasis
Meta executives' statements, the Scale AI investment, AI spending, the shift away from Llama and cost.
Leaves out or plays down
Does not mention the access tiers or API availability, or the new agent features.
Charged language
“so-called benchmarking data”“rivals the big (boys)”
For example
“Meta touted so-called benchmarking data from an independent firm called Artificial Analysis” — The Straits Times
“Meta says its new AI models on par with OpenAI, Anthropic” — The Straits Times

Ask about this story

Loading…