Meta has introduced two new artificial intelligence models, Muse Wonder and Muse Wonder Full, positioning them as its most capable systems to date and a step toward "personal superintelligence." Announced on Monday, the models feature two key capabilities: Active Memory, which retains relevant information across conversations, and Active Focus, designed to help agents sustain work on complex tasks. According to independent evaluation platform Artificial Analysis, Muse Wonder Full scored 56 on its Intelligence Index at maximum reasoning effort, while the standard Muse Wonder scored 52. These results place Full above Claude Fable 5.1 and GPT-6 Astra, both of which scored 53, though below Claude Opus 5.5’s score of 58.
The pricing structure for the new models reflects their advanced capabilities. The standard Muse Wonder charges $4 per million input tokens, matching the cost of Claude Opus 5.5 but significantly undercutting GPT-6 Astra ($10) and Claude Fable 5.1 ($50). Despite Meta's claims, Bindu Reddy, CEO of Abacus.AI, challenged the performance assertions on X, stating that Muse Wonder was "nowhere near Fable or Astra level" and indicating that detailed benchmark results would follow.
Meta's launch of Muse Wonder and Muse Wonder Full signals an aggressive push into high-performance AI infrastructure, leveraging competitive pricing to challenge established leaders like OpenAI and Anthropic. By introducing features such as Active Memory and Active Focus, Meta aims to differentiate its offerings through sustained contextual awareness and task persistence, addressing common limitations in current large language models. The strategic pricing, particularly the $4 per million token rate for the standard model, suggests an intent to capture market share by offering comparable or superior performance metrics at a lower cost than some competitors.
However, the immediate skepticism from industry figures like Bindu Reddy highlights the contentious nature of AI benchmarking and performance claims. While Artificial Analysis provides third-party validation, discrepancies between vendor claims and independent expert assessments underscore the complexity of evaluating model capabilities. This dynamic creates operational risk for enterprises considering adoption, as they must navigate conflicting data points regarding reliability and comparative advantage. The outcome will likely depend on real-world deployment consistency rather than isolated index scores.
