Meta Launched Muse Wonder: Two Frontier Models for the Agent War (With Aggressive Pricing) | inicIA
Direct answer
On October 12, 2026, Mark Zuckerberg announced Muse Wonder, the new family of frontier models from Meta Superintelligence Labs (MSL), led by Alexandr Wang. Two models built for long-haul agentic work: Muse Wonder, available on the Meta Model API and in Muse Code, and Muse Wonder Full, the complete version that lives only inside Muse Code ($50 and $200-per-month plans). The price is what moved the conversation: $4 per million input tokens and $20 output — identical to Claude Opus 5.5, its direct rival. Meta describes it as its most capable model yet, with two agent-focused capabilities: Active Memory (memory across conversations) and Active Focus (sustaining effort on difficult goals).
The launch facts
- Date and announcement: October 12, 2026, presented by Zuckerberg as "the next major step toward personal superintelligence for everyone."
- Two models, two doors: Wonder (standard model, API + Muse Code) and Wonder Full (maximum reasoning for the longest, hardest work, Muse Code only).
- Pricing: $4 input / $20 output / $0.20 cached per million tokens on the API — the same rate as Claude Opus 5.5, per launch coverage.
- Benchmarks: on the independent Artificial Analysis Intelligence Index, Claude Opus 5.5 scores 58, Wonder Full 56, and Wonder 52. On Meta's own measurements (SWE-Marathon 57.5 vs Opus 5.5's 46.3; DeepSWE 1.1 79.0 vs 74.2), Wonder Full wins.
- Training: Meta says they were trained at its gigawatt-scale Prometheus data center in Ohio.
- No open weights: a break from Meta's Llama lineage; the models are proprietary.
Active Memory and Active Focus: what they solve
Meta designed Wonder for tasks that stretch over time — reading a lease, processing an insurance claim, writing a complete application. Active Memory carries relevant context across conversations (days or weeks) and retrieves what matters for the task. Active Focus sustains effort toward the stated goal: it tries alternative approaches when the first one fails, asks for clarification instead of guessing, and checks in before important actions. They attack two typical agent failure modes: losing context in long sessions and drifting from the original objective.
What it means for your business
- Price became the battlefield: three competitors (Meta, Anthropic, OpenAI) now charge close rates in the premium tier. Cost per task — not per token — is the metric that matters when choosing.
- The access route matters as much as the model: Wonder Full, the most capable, exists only inside a subscription. Before building on a model, verify where you can use it, not just its score.
- Persistent memory is the new differentiator: for agents working business processes over days, the ability to hold context is worth more than a few benchmark points.
Related links
- OpenAI Started Invisibly Watermarking ChatGPT Text: What It Means for Your Business
- Claude Haiku 5.5 Costs Up to 90% Less: The AI Price War Continues (and Benefits You)
- Small-Business AI Spending Skyrocketed and Anthropic Took the Crown from OpenAI: Your Buyer Advantage
Frequently asked questions
What is Muse Wonder?
Meta's new family of frontier models (Oct 12, 2026), built by Meta Superintelligence Labs for personal and professional agentic work: long tasks with multiple steps that require planning, tool use, and context retention.
What is the difference between Wonder and Wonder Full?
Wonder is the standard version, available on the Meta Model API and in Muse Code. Wonder Full is the complete model, running at maximum reasoning, available only to Muse Code subscribers.
How much does Muse Wonder cost?
On the API: $4 per million input tokens and $20 output ($0.20 cached) — the same rate as Claude Opus 5.5. Wonder Full has no API price: it is bundled with Muse Code subscriptions ($50 and $200 per month, per coverage).
Did Meta release the model weights?
Per launch coverage, neither the announcement nor the model card mentions open weights. It is a proprietary release, unlike Meta's historic Llama lineage.
Take the next step
Choosing the right model for your business — cost per task, access, and persistent memory — is a technical decision with direct impact on costs. If you want that evaluation done by experts, talk to inicIA on WhatsApp: https://wa.me/17866120930
Chat on WhatsAppBy inicIA
