What Is Muse Wonder? Meta's Frontier Agent Model

Inside Muse Wonder: The 56-Point Climb to #2 on the Artificial Analysis Index

Muse Wonder is a frontier large language model developed by Meta's Superintelligence Labs, released on October 12, 2026, for personal and professional agentic tasks. It ships in two variants — Muse Wonder and Muse Wonder Full — and Muse Wonder Full scored 56 on the Artificial Analysis Intelligence Index, placing it second in the world behind Claude Opus 5.5 at 58. Muse Wonder is available through the Meta Model API and Muse Code, with the more capable Full variant reserved for Muse Code subscribers.

Key Takeaways

• Muse Wonder and Muse Wonder Full are Meta's most capable models to date, launched October 12, 2026, and positioned for agentic work.
• Muse Wonder Full scored 56 on the Artificial Analysis Intelligence Index v4.3; Muse Wonder scored 52, up from 48 for Muse Spark 1.3 six weeks earlier.
• At 56, Muse Wonder Full sits two points behind Claude Opus 5.5 (58) and ahead of GPT-6 Astra and Gemini 4 Argon.
• Reported API pricing is $4 per million input tokens, $20 output, and $0.20 cached input — matching Opus 5.5's headline rate — though Meta has not officially confirmed these figures.
• The models are built around two named features: Active Memory (retaining relevant context across conversations) and Active Focus (sustaining effort on long-running tasks).
• Muse Wonder is on the Meta Model API and Muse Code; Muse Wonder Full is subscriber-only and not on the API at launch.

What Is Muse Wonder?

Muse Wonder is Meta's frontier model series, built by its Superintelligence Labs (MSL) and announced publicly by Alexandr Wang on October 12, 2026. Meta describes the series as its "most capable models" for personal and professional agentic tasks, framing the release as "another step toward bringing personal superintelligence to the billions of people who use our products."

The series has two members. Muse Wonder is the standard variant, described by Meta as "state-of-the-art at real-world coding, built for long-running tasks." Muse Wonder Full runs at maximum reasoning for the hardest, longest-running jobs. According to Meta's research blog, the models were trained at a gigawatt-scale data center Meta calls Prometheus, and the release builds on the pretraining and reinforcement-learning approach introduced with Muse Spark 1.3 in April 2026.

This is a confirmed, shipped release rather than a leak. The first endpoint sighting came roughly 30 minutes before the official announcement, when community watchers spotted "muse-wonder-1.0" on the Meta Models API.

Source: @Mr_Salio

Muse Wonder at a Glance

Attribute	Detail
Developer	Meta (Superintelligence Labs / MSL)
Type	Frontier LLM, agentic / coding focus
Variants	Muse Wonder, Muse Wonder Full
Modality	Text, image, audio, video inputs (reported, unconfirmed)
Context window	1M tokens (reported by a single source, unconfirmed)
Pricing (Wonder)	$4 input / $20 output / $0.20 cached per 1M tokens (reported, not officially confirmed)
Pricing (Wonder Full)	No API price; subscriber-only
Availability	Meta Model API, Muse Code (Wonder); Muse Code subscribers (Wonder Full)
License	Proprietary
Released	October 12, 2026

How Muse Wonder Works / What Makes It Different

Meta anchors the Muse Wonder series on two named capabilities. Active Memory is how the model carries important context about a user across conversations and retrieves the details relevant to the task at hand. Active Focus describes how it sustains effort toward a difficult goal — trying alternative approaches, working through blockers, and adapting to new information as it arrives.

Those two ideas sit under a larger program Meta calls Personal Superintelligence, which the company uses to describe its scaling roadmap. The standard Muse Wonder variant is tuned for agentic coding: Meta says it "plans the work, calls tools in parallel, hands parts off to sub-agents, and recovers from errors so complex jobs finish." Domain expertise is a second emphasis, with evaluations spanning law, finance, and health — including Harvey's Legal Agent Benchmark (LAB).

The benchmark story is the reason this launch moved markets of attention. On the Artificial Analysis Intelligence Index v4.3, Muse Wonder Full scored 56 and Muse Wonder scored 52, against 48 for the prior Muse Spark 1.3. A six-week gap between a 48 and a 56 is a steep curve for one vendor.

Source: @ChrisGPT

One caution runs through the community reaction: most of the headline coding numbers are Meta's own. Several observers flagged possible benchmark optimization, and one widely shared comparison explicitly noted that the evaluation setups between vendors "aren't identical." Treat the cross-vendor coding deltas as directional until independent runs land.

What You Can Do With Muse Wonder

The intended workloads cluster around long-running agent tasks. Meta positions the standard model for real-world coding — the kind of job that plans across many steps, spawns sub-agents, and has to recover from its own mistakes before finishing. Early users report building inside Muse Code with Muse Wonder Full as the default engine.

The domain-expertise framing points at knowledge work: reading a lease, making sense of financial statements, or interpreting health information. One analyst summarized the targeting as "knowledge work, health, finance, legal," backed by specialized datasets.

Usage reports are early and mixed. One user said a single Muse Wonder Full run exhausted a five-hour limit and consumed 38% of a weekly allowance, which suggests the Full variant is compute-heavy on long jobs. For teams evaluating a comparable frontier chat-and-agent engine today, Claude Opus 5.5 is the model Muse Wonder is most often measured against, and it is accessible through standard API channels.

How Muse Wonder Compares

Reported, Meta-sourced figures place the models as follows. Numbers marked as reported come from launch-day posts, not independent testing.

Benchmark	Muse Wonder Full	Claude Opus 5.5	Gemini 4 Argon
Artificial Analysis Index v4.3	56	58	53 (reported)
DeepSWE 1.1	79%	74.2 (reported)	77.9%
SWE-Marathon	57.5%	46.3% (reported)	Not yet confirmed
ProgramBench	50.0 (reported)	70.5 (reported)	Not yet confirmed
Terminal-Bench	57.1%	59.6 (reported)	57.4%

The pattern: Muse Wonder Full leads on several agentic and long-horizon coding evaluations while trailing Opus 5.5 on ProgramBench and on the overall Index. On pricing, the reported $4/$20 rate matches Opus 5.5's headline token price, a parallel we covered in our look at Claude Opus 5.5's token-price signal. Against Google, one comparison argued Gemini 4 Argon was outscored on the Index and DeepSWE but still led on AutomationBench (51.3% vs 45.4%); for context on that model see our Gemini 4 Argon overview.

Availability: How to Access Muse Wonder

Muse Wonder is available through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers and, as of launch, is not offered on the API. Meta says Muse Wonder "will soon power experiences in Muse" and roll out first to subscribers, later driving artifact generation and higher-intelligence processes.

The official announcement came from Alexandr Wang and Meta's channels; tier-1 outlets including CNBC and Bloomberg reported the launch the same day. Full technical detail lives on Meta's AI research blog, which hosts the Muse Wonder model page and evaluation report.

What We Don't Know Yet

Several load-bearing specs remain unconfirmed by Meta:

• Official pricing. The $4/$20/$0.20 figures appear across multiple non-official posts but were not stated in an official Meta post in the sources reviewed. Which variant they apply to is also unclear.
• Context window and modality. A 1M-token window with text, image, audio, and video inputs is a single-source report, not a confirmed spec, and it is unclear whether it applies to both variants.
• Wonder Full API access. Whether the Full variant ever gets a public API and a published price is unresolved.
• Benchmark generalization. Most coding results are Meta's own numbers; independent evaluations had not landed at the time of writing, and at least one comparison noted non-identical setups.
• The Muse "Maximum" plan. A reported $100 plan with 3 billion weekly tokens for "latest models" is a single-source attribution to Muse Wonder.

Frequently Asked Questions

What is Muse Wonder?

Muse Wonder is a frontier large language model developed by Meta's Superintelligence Labs, released on October 12, 2026, for personal and professional agentic tasks. It ships in two variants, Muse Wonder and Muse Wonder Full, and is built around features Meta calls Active Memory and Active Focus.

Who made Muse Wonder?

Meta made Muse Wonder through its Superintelligence Labs (MSL), and the launch was announced by Alexandr Wang. Meta says the model is the result of more than a year of research and infrastructure work, trained at its gigawatt-scale Prometheus data center.

How much does Muse Wonder cost?

Muse Wonder is reported to cost $4 per million input tokens, $20 per million output tokens, and $0.20 per million cached input tokens, according to multiple non-official posts. Meta had not published an official API price confirmation in the sources reviewed.

Is Muse Wonder better than Claude Opus 5.5?

Muse Wonder Full scored 56 on the Artificial Analysis Intelligence Index versus 58 for Claude Opus 5.5, so it sits two points behind on that benchmark. Meta-reported figures show Muse Wonder Full ahead on some coding evaluations such as DeepSWE 1.1 and SWE-Marathon, while Opus 5.5 leads on ProgramBench.

What is the difference between Muse Wonder and Muse Wonder Full?

Muse Wonder is the standard variant available through the Meta Model API and Muse Code, while Muse Wonder Full is Meta's most capable variant running at maximum reasoning and is available only to Muse Code subscribers. Muse Wonder Full is not offered on the API as of launch.

Where can I access Muse Wonder?

Muse Wonder is available through the Meta Model API and Muse Code, and Muse Wonder Full is available to Muse Code subscribers. Meta says Muse Wonder will soon power experiences in its Muse product, rolling out first to subscribers.

What context window does Muse Wonder have?

One post reported a 1 million-token context window with text, image, audio, and video inputs for Muse Wonder and Muse Wonder Full. This figure is single-source and unconfirmed by Meta in the sources reviewed.

What to Watch Next

Three signals will tell you whether the launch-day numbers hold. First, watch for the independent Artificial Analysis writeup — Meta posted its figures ahead of the public index, so the third-party confirmation is still pending. Second, watch for an official pricing page that either confirms or revises the reported $4/$20/$0.20 rate and clarifies whether Muse Wonder Full ever reaches the API. Third, watch independent coding benchmarks, since the current cross-vendor deltas rest on Meta's own evaluations. This page will be updated as those land.

Try related models on Kie.ai: Claude Opus 5.5, GPT-6 Astra and Claude Fable 5.1 (Coming Soon).