Key Points
- Meta Platforms describes its new Muse Wonder models as frontier-class, a major shift from its earlier, more cautious framing of Muse Spark as merely competitive.
- Third-party evaluator Artificial Analysis scored Muse Wonder Full at 56, outperforming OpenAI's GPT-6 Astra and Alphabet's Gemini 4 Argon.
- Meta raised Muse Wonder's per-task pricing and locked the more powerful Muse Wonder Full behind paid Muse Code subscriptions ahead of its Oct. 21 earnings release.
- Special Report: Ready to give options a try? Your first trade (Ticker included) -INSIDE
Magnificent Seven company Meta Platforms (NASDAQ: META) recently made a landmark statement related to its artificial intelligence journey.
The company claims that Muse Wonder, its latest model family, is frontier-class, putting it in the same territory as models developed by OpenAI and Anthropic.
This marks the first time Meta has framed its Muse models as among the most intelligent available. And, investors don’t just have to take Meta’s word for it; model evaluation sites provide third-party validation of this claim.
As Meta looks to improve AI monetization over time, developing a leading-class model could be a key benefit the firm has historically lacked.
Meta’s New Muse Models Move Up in Frontier Ranking Hierarchy
Approximately 14 months ago, Meta began to “overhaul” its AI efforts. This overhaul included launching Meta Superintelligence Labs (MSL) and hiring Alexandr Wang to lead the division. Before this, investors saw Meta’s Llama models as lagging far behind the frontier, leaving them with limited confidence in the company’s ability to compete with other LLM developers.
By April 2026, MSL released its first model, Muse Spark. Although the firm noted that Muse Spark was its most powerful model yet, the company framed it as being “competitive” rather than pushing the frontier of model development. With the release of Muse Wonder and Muse Wonder Full, that framing has changed.
Meta now notes that the Muse Wonder models “deliver state-of-the-art performance on personal and professional agentic tasks.” Data from the model evaluation site, Artificial Analysis, supports Meta’s assessment and highlights a significant jump in performance compared to its previous model, Muse Spark 1.3.
On its Artificial Analysis Intelligence Index, the evaluator gives Muse Wonder Full a score of 56. This ties for the second-highest score of any model, with Anthropic’s Claude Sonnet 5.5 also scoring 56. Anthropic’s Claude Opus 5.5 has the highest score at 58. Muse Wonder Full notably ranks above OpenAI’s GPT-6 Astra and Alphabet’s (NASDAQ: GOOGL) Gemini 4 Argon, which both have scores of 53. Meanwhile, Muse Wonder scores slightly lower at 52.
With Muse Wonder Full and Muse Wonder both ranking in this upper echelon, Artificial Analysis’s data supports the idea that Meta now has frontier-class models in its hands. The firm also notes a significant jump in the capabilities of these models compared to Muse Spark 1.3, which has a score of 48.
Meta Raises Model Prices, Keeps Muse Wonder Full Behind Subscription Wall
Interestingly, Muse Wonder comes with a hefty price increase versus Muse Spark 1.3. On their max settings, Muse Wonder’s cost per Artificial Analysis Intelligence Index task is $4.52, or 2.8 times higher than Muse Spark 1.3’s cost of $1.60. This is still meaningfully below the cost per task of Anthropic’s most intelligent models, but also significantly higher than OpenAI and Alphabet’s top models. The increase suggests that Meta believes the model’s improved capabilities justify a higher price.
On the other hand, the more powerful Muse Wonder Full is only available through $50 per month and $200 per month subscriptions to Muse Code, not on a per-token basis. This suggests that Meta primarily wants to position Muse Wonder Full as being for agentic coding tasks, while Muse Wonder is more for general-purpose usage. It also suggests that Meta hopes to build a more stable, subscription-based revenue stream by keeping Muse Wonder Full behind this gate.
Additionally, by keeping Muse Wonder Full access to only Muse Code subscribers, it gives software developers a reason to try Muse Code. This could help Meta position Muse Code as a competitor to Anthropic’s Claude Code and OpenAI’s Codex. Still, only time will tell whether software developers actually stick with Muse Code rather than experimenting with it for a short time.
Meta notes that Muse Wonder will soon power more of its Muse agent. As the company looks to scale this product, Muse Wonder’s improved capabilities could improve user experiences and aid demand.
Meta Releases a Flurry of Products With Q3 Earnings on the Horizon
Meta has demonstrated the ability to significantly improve its models in a relatively short time. This bodes well for the company’s ability to continue competing with other model developers on frontier intelligence and to scale its non-advertising AI revenue. However, up to this point, Meta’s non-ad AI potential has yet to meaningfully translate into its financial results.
The Muse agent, the Meta Enterprise Platform, Muse Code, and now the Muse Wonder model releases should make Meta’s future earnings calls more interesting. Investors should look for signs of strong early adoption and, over time, a contribution to revenue growth. Meta’s next earnings report is quickly approaching, slated for Oct. 21 after the market close.
Read this article online ›
The best investment opportunities don't wait. Get our research and stock ideas delivered straight to your smartphone—so you never miss a market-moving opportunity. Our text alerts ensure you see timely stock ideas and professional research reports instantly, whether you're in a meeting, commuting, or away from your desk.
Get Text Alerts from American Market News (free)
Comments
Post a Comment