Key Highlights:
- The new Full model achieved a benchmark score outperforming Claude Fable 5.1 and GPT-6 Astra (both at 53), while trailing Claude Opus 5.5’s score of 58.
- Abacus.AI CEO Bindu Reddy openly questioned the benchmark claims, stating that Muse Wonder was “nowhere near Fable or Astra level” ahead of planned independent evaluation releases.
- Meta’s standard Wonder model is priced at $4 per million input tokens, whereas the top-tier Wonder Full access is restricted to Muse Code subscriptions costing $50 or $200 per month.
Performance Benchmarks Place Wonder Between Leading Rivals
In a newly released round of benchmark evaluations, the Full model has established a competitive footprint among top-tier artificial intelligence offerings. Scoring higher than both Claude Fable 5.1 and GPT-6 Astra—each of which registered a score of 53—Full positioned itself as a major contender. However, it still falls short of Claude Opus 5.5, which leads the bracket with a score of 58.
Industry Scrutiny and External Evaluations
Despite the published benchmark numbers, external experts have raised doubts regarding the actual capabilities of the architecture. Bindu Reddy, CEO of Abacus.AI—an organization known for running independent model evaluations—strongly pushed back against the performance metrics. Taking to X, Bindu Reddy argued that Muse Wonder was “nowhere near Fable or Astra level”
and indicated that comprehensive benchmark evaluations would be released soon to provide an independent perspective.
Pricing Architecture and Tiered Access
The latest rollout marks Meta’s most expensive model release to date. Under its current structure, the standard Wonder model costs $4 per million input tokens. This token pricing matches the rate charged for Claude Opus 5.5 and significantly undercuts the higher token costs of GPT-6 Astra and Claude Fable 5.1, which demand $10 and $50 per million input tokens, respectively.
For users seeking the highest-performing capabilities, deployment of Wonder Full has been gated behind premium service tiers. Access to Wonder Full is currently limited exclusively to users who hold subscriptions to Muse Code’s $50 per month “power usage”
tier or the $200 per month “maximum usage”
tier.
Why This Matters
The release reflects an aggressive push to balance elite performance metrics with competitive token pricing against dominant ecosystem rivals like the Claude and GPT families. By setting token fees at $4 per million input tokens for the standard model while locking Wonder Full behind high-cost $50 and $200 developer tiers, Meta is actively experimenting with dual-track monetization. However, the immediate public pushback from figures like Bindu Reddy highlights that as AI frontier models crowd the upper scoring brackets, verified third-party evaluations are becoming critical for validating real-world enterprise utility.
Frequently Asked Questions
How did the Full model score compared to other leading AI models?
The Full model recorded a benchmark score higher than Claude Fable 5.1 and GPT-6 Astra (both scoring 53), but placed behind Claude Opus 5.5, which scored 58.
What does Meta charge to use the standard Wonder model?
Meta charges $4 per million input tokens for the standard Wonder model, placing it on par with Claude Opus 5.5 and well below the $10 charged for GPT-6 Astra and $50 charged for Claude Fable 5.1.
How can developers access the Wonder Full model?
Access to Wonder Full is restricted to subscribers of Muse Code, requiring enrollment in either the $50-per-month “power usage” plan or the $200-per-month “maximum usage” tier.
