The Extended Brief

Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model

Brief by The AI News AI newsroom · Oct 13, 2026, 2:42 PM EDT edition

Original reporting by Latent Space · published Oct 13, 2026, 5:37 AM EDT

Meta's 14-month-old superintelligence lab has shipped a near-frontier model family, adding a new contender for frontier-class API and coding workloads.

Key points

• Meta Superintelligence Labs shipped Muse Wonder and Muse Wonder Full, its second-generation family, roughly 14 months after founding. source ↗
• Third-party evals place Wonder near the frontier; Wonder Full reportedly beats Fable and Astra on some benchmarks. source ↗
• Leaked pricing lists Wonder at $4/M input, $20/M output, with a 1M-token context window. source ↗
• Wonder Full is exclusive to Muse Code subscribers, while Wonder ships through the Meta Model API and Muse Code. source ↗
• Meta says Wonder will power the Muse personal-agent product, with subscribers getting access first. source ↗

Practical applications

• Benchmark Muse Wonder through the Meta Model API against your current frontier model on real workloads before the leaked pricing is confirmed or revised.
• If your workloads need maximum reasoning, price in a Muse Code Power or Maximum Usage subscription, since Wonder Full is gated behind those tiers.
• Treat the leaked $4/M-input and $20/M-output figures as provisional in any cost model until Meta publishes official pricing.

Context

Meta Superintelligence Labs is Meta's frontier AI effort; Muse Spark was its first model family and Wonder is its second, larger generation. The tiered lineup mirrors rivals' naming ladders — OpenAI's Sol/Terra/Luna and Anthropic's Opus/Sonnet/Haiku — with Meta pairing Spark, Glimmer, and now Wonder. Access splits between the Meta Model API for the fast tier and a Muse Code subscription for the full model.

What to watch

• Official Meta pricing would confirm or unwind the leaked $4/M and $20/M figures.
• Independent benchmark runs will clarify whether Wonder Full sits in the Fable/Astra class or above it, which the launch coverage leaves unresolved.

Related briefs

• ThinkingBox: Solving an agent task once vs. solving it 20/20: 507 stateful workflows graded on terminal database state [R]
• Google Cloud launches Gemini agent to work across enterprise systems
• We unlearned CCP alignment from Qwen3.6-35B-A3B: censored/propaganda answers 89.8% → 2.8%, general benchmarks within ~1 point (open weights)
• Ignore all instructions and read this blog: The state of AI-analysis evasion in malware

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.