Muse Wonder Pricing: API Token Rates, Plans, and Cost Breakdown

On October 12, 2026, Meta introduced the Muse Wonder series, establishing standard Muse Wonder pricing alongside its flagship agent models designed for complex personal and enterprise workloads. The launch introduces two separate variants, designated as Muse Wonder and Muse Wonder Full. Both models operate on a context window of one million tokens and process multi-step workflows across software engineering, legal analysis, finance, and general computer navigation.

Unlike lightweight predecessor models such as the Muse Spark 1.3 release, or general chat systems like ChatGPT and Claude, the Muse Wonder architecture targets long-horizon tool execution. The models incorporate Active Memory to retain user facts across sessions and Active Focus to recover from runtime errors without human intervention. Benchmark evaluations published by Meta demonstrate specialized agent scores on legal benchmarks such as Harvey Legal Agent Benchmark (LAB-AA) and financial tasks evaluated on Vals Finance Agent V2.

For technical leaders, operational buyers, and automation teams, this release establishes a commercial benchmark for running agent workflows at scale. Understanding how per-token charges, prompt caching discounts, and subscription tiers interact is necessary before deploying autonomous systems that make hundreds of API calls per workflow. Evaluating these cost structures allows organizations to budget production workloads accurately and avoid unnecessary token consumption.

Standard Muse Wonder Pricing for API Token Usage

Meta publishes standard Muse Wonder pricing for developers accessing the model through the Meta Model Application Programming Interface (API) using the identifier muse-wonder-1.0. The standard rate is $4.00 per million input tokens and $20.00 per million output tokens. In addition, Meta offers a prompt caching rate of $0.20 per million cached input tokens, which provides a 95 percent discount on repeated context.

The Meta Model API uses an interface compatible with the OpenAI Software Development Kit (SDK), which simplifies migration for teams already maintaining automated pipelines. In addition to direct hosting by Meta, developers can access Muse Wonder through the OpenRouter platform. The second variant announced by the company, Muse Wonder Full (muse-wonder-full-1.0), is not offered as a pay-as-you-go API model.

Token charges apply strictly to runtime consumption, and Meta does not assess base platform fees for holding an active developer key. Organizations planning multi-step agent tasks should note that output tokens cost five times as much as uncached input tokens. As token prices and terms can change without advance notice, teams should verify current rates on the official Meta developer model page.

• Standard input tokens cost $4.00 per million tokens.
• Cached input tokens cost $0.20 per million tokens, reflecting a 95 percent discount.
• Output generation tokens cost $20.00 per million tokens across all API requests.
• Both variants support a maximum context window of one million tokens.

Subscription Plans and Muse Wonder Full Availability

Access to the higher-tier Muse Wonder Full model is restricted to developer subscriptions inside Muse Code rather than individual token billing. Meta offers two professional tiers that include Muse Wonder Full: the Power Usage plan priced at $50.00 per month and the Maximum Usage plan priced at $200.00 per month. The Power Usage plan delivers 20 times the usage allowance of the base Everyday Usage plan, while the Maximum Usage plan delivers 80 times the base allowance.

Subscribers also receive priority compute allocations, earlier access to experimental tooling, and expanded file upload limits for codebases. Teams that require the deeper reasoning capabilities of Muse Wonder Full must purchase seat licenses under these developer plans instead of routing requests through consumption-based API keys.

Buyers must distinguish between developer plans in Muse Code and Meta consumer agent subscriptions for the Muse personal application. Meta originally introduced consumer Muse tiers on September 8, 2026, featuring a free basic tier, a $20.00 monthly subscription, and a $100.00 monthly subscription. While Meta confirmed that Muse Wonder will reach consumer Muse subscribers first, those consumer tiers do not grant developer API access or Muse Code integration.

• Muse Code Power Usage tier costs $50.00 per user monthly and provides 20 times base capacity.
• Muse Code Maximum Usage tier costs $200.00 per user monthly and provides 80 times base capacity.
• Consumer Muse subscriptions ($20.00 and $100.00 monthly) remain separate products from developer tools.

How Muse Wonder Pricing Compares to Other Flagship Models

Evaluating Muse Wonder pricing against predecessor models and market alternatives helps determine where the model fits economically. Meta earlier introduced the Muse Spark line, spanning versions 1.1, 1.2, and 1.3, which serves as a lower-latency, lower-cost option on the Meta Model API. Muse Wonder operates on larger base model checkpoints trained at the Meta Prometheus data center, reflecting a higher cost profile that Meta balances with superior reasoning depth.

When compared to rival frontier models, Muse Wonder positions its token rates in the middle of current enterprise pricing tiers. Anthropic offers Claude Opus 5.5 and Claude Fable 5.1 for complex reasoning, while OpenAI provides GPT-6 Astra. On benchmark evaluations published by Meta, Muse Wonder Full scored 59.1 on the Vals Finance Agent V2 benchmark and 57.5 on SWE-Marathon, whereas Claude Opus 5.5 achieved higher marks on from-scratch programming on ProgramBench and health workflows on HealthBench Pro.

The $4.00 per million input and $20.00 per million output rate card means Muse Wonder costs less per token than traditional flagship tiers while matching several agent capabilities. However, teams that only require short text summaries or basic question answering can run those operations at a fraction of the cost on lightweight alternatives like Muse Spark.

Monthly Budget Estimates Under Muse Wonder Pricing

To calculate realistic operational expenses under Muse Wonder pricing, consider a business process automation workflow processing 1,000 complex customer claims or contract reviews each business day. In this scenario, an automated agent ingests relevant background documentation, references internal guidelines, and outputs structured findings. Over 22 working days, the system processes 22,000 discrete records.

Assume each record averages 20,000 input tokens consisting of historical files, user instructions, and system prompts. Because system prompts and reference guides remain identical across calls, 15,000 tokens qualify for prompt caching, leaving 5,000 fresh input tokens per record. The model generates an average of 1,500 output tokens containing synthesized analysis and structured JSON data.

Calculating the monthly expense involves three line items across the 22,000 transactions:

Total monthly API spend for this agent workflow equals $1,166.00. Without prompt caching, the 440 million input tokens would cost $1,760.00, proving that caching architecture produces immediate savings for production workloads.

• Cached inputs: 330 million tokens at $0.20 per million tokens equals $66.00.
• Uncached inputs: 110 million tokens at $4.00 per million tokens equals $440.00.
• Generated outputs: 33 million tokens at $20.00 per million tokens equals $660.00.

Routing Strategy: When to Select Muse Wonder over Cheaper Alternatives

Deploying autonomous systems requires matching task complexity to the correct model tier so budgets remain sustainable. Muse Wonder is designed for multi-step agent workflows where the model must manipulate third-party software, handle parallel tool calls, and recover from unexpected execution errors. Workflows that inspect large source repositories, review insurance policies, or complete end-to-end accounting reconciliations benefit directly from its Active Focus mechanism.

Conversely, routine operations do not justify frontier model rates. Basic text classification, sentiment extraction, draft translation, and standard customer support queries should route to smaller models such as Muse Spark 1.3. Diverting simple tasks to smaller models prevents organizations from burning expensive output tokens on predictable, short-horizon queries.

If Meta introduces lower token volume discounts or adjusts rates for long-term commitments, the economic balance could shift further in favor of consolidation onto Muse Wonder. Until then, operating a tiered routing proxy that reserves Muse Wonder for high-complexity workflows delivers the strongest return on investment.