Meta Models: Every Model, Spec, Price and Release Date

Meta develops the Muse family of large language models, built by Meta Superintelligence Labs, alongside its earlier open-weight Llama models, and offers them through the Meta Model API, Meta AI apps and open-weight downloads. This hub lists the Meta models in Meta's public documentation as of 15 November 2026, with API IDs, release dates, context windows, prices, availability, licences and access conditions. It is written for developers, buyers, architects and anyone who needs to know which Meta model is which. Models are listed by release date, newest first, an objective order that implies no ranking; models whose release date is not stated on the pages checked are listed last. Every fact comes from Meta's own documentation, pricing page, announcements or Hugging Face model cards, cited in the sources list.

Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. It uses Meta's public model documentation as of November 2026. Model details change, so confirm against Meta's current documentation.

Key takeaways

• Three Muse lines are on the Meta Model API. Muse Wonder (muse-wonder-1.0, in preview), Muse Spark (1.1, 1.2 and 1.3, with 1.3 recommended for new work) and Muse Ember (muse-ember-1.0).
• Two Muse models have open weights. Muse Spark Open (816B total, 42B active parameters, Muse Model License, released 15 October 2026) and Muse Glimmer 30B (Apache 2.0, August 2026).
• API prices run from $0.10 to $4 per million input tokens. Muse Ember starts at $0.10 input and $0.50 output; Muse Spark is $1.25 and $4.25; Muse Wonder is $4 and $20.
• Contributor mode trades data for price. Contributor variants of Muse Spark 1.3 and 1.2 cost $0.10 input and $0.20 output in exchange for permission to use the customer's content to train future Meta models.
• Llama 4 remains the latest Llama release. Llama 4 Scout (109B total parameters, 10M context) and Maverick (400B total, 1M context), both with 17B active parameters, are on Hugging Face under the Llama 4 Community License.

What is Meta?

Meta Platforms builds AI models through Meta Superintelligence Labs, which introduced Muse Spark, the first Muse model, on 8 April 2026. Meta's model documentation lists Muse Wonder, Muse Spark and Muse Ember on the Meta Model API, plus Muse Image, Muse Voice Transcribe and the Segment Anything Model (SAM 3.1). Muse Spark Open and Muse Glimmer are open-weight models with no Model API ID; they are self-hosted or used through hosted providers. The earlier Llama family, last updated with Llama 4 in April 2025, is documented on Meta's Llama pages.

The Muse API models are proprietary, while Muse Spark Open, Muse Glimmer and Llama 4 publish weights under the licences set out below. Meta states the live catalogue, including created timestamps, is available from the models list endpoint of the Model API. Facts are as of 15 November 2026.

Current Meta models at a glance

As of 2026-11-15. Source: Meta Model API models page, pricing page, announcements and Hugging Face model cards. On this page, "Not stated by Meta" means not stated by Meta on the pages checked (2026-11-15). Category is the Meta model line.

Llama 4 rows are placed by their April 2025 release; Muse Spark 1.3, 1.2 and Muse Ember have no release date on the pages checked and are listed last. See Open-Weight vs Closed Models Explained.

Key facts for the main Meta models

As of 2026-11-15. Identical fields for every model.

The Llama 4 release date is from the Scout model card on Hugging Face; the Maverick row uses the same Llama 4 launch. The Hugging Face metadata for Muse Spark Open lists 822B parameters, while the card text states 816B total.

Features stated across current models

• Modes: Muse Spark has a Contemplating mode, a multi-agent reasoning mode rolled out on meta.ai, and a Thinking mode in the Meta AI app.
• Agentic features: Muse Wonder adds Active Memory, which carries context across conversations, and Active Focus, which keeps pursuing a goal and asks for clarification when needed.
• Grounding: web search grounding is billed at $2.50 per 1,000 queries on top of token costs.
• Other API models: Muse Image (generation and editing), Muse Voice Transcribe (speech to text) and SAM 3.1 (segmentation boxes and masks).

Meta model families and naming

• Muse Wonder: Meta's agentic frontier line, building on Muse Spark. A Muse Wonder Full variant is available only to Muse Code subscribers.
• Muse Spark: the first Muse line, versioned 1.0 to 1.3 with API IDs such as muse-spark-1.3. Contributor variants add -contributor to the ID.
• Muse Ember: the lowest-priced Muse API model, with a price tier at 200K input tokens.
• Muse Spark Open: an open-weight Muse model named with its parameter counts (816B-A42B: 816B total, 42B active).
• Muse Glimmer: an open-weight 30B model distilled from Muse Spark, for self-hosting, with quantised builds.
• Llama: Meta's earlier open-weight family. Llama 4 IDs encode active parameters and expert count, for example Llama-4-Maverick-17B-128E. See LLM Naming and Versioning Explained.

When was each Meta model released?

As of 2026-11-15. Dates are from Meta announcements and Hugging Face model cards, in date order.

The latest dated release on the pages checked is Muse Spark Open, on 15 October 2026. Muse Spark 1.2, Muse Spark 1.3 and Muse Ember are listed on the models and pricing pages without a release date.

How much do Meta models cost?

As of 2026-11-15. Price per million tokens on the Meta Model API, pay-as-you-go. The pricing page does not state a currency or batch prices on the content fetched; amounts use the dollar sign as shown. Source: Meta Model API pricing and rate limits.

• Non-token models: Muse Image costs $0.01 per generated image; Muse Voice Transcribe $0.18 per hour of audio; SAM 3.1 $2.50 per 1,000 images or $0.20 per 1,000 video frames.
• Rate limits: Muse Wonder, Muse Spark and Muse Ember allow 3,000 requests and 4,000,000 tokens per minute per team; contributor variants allow 150 requests and 3,000,000 tokens per minute. Limits apply per team, not per API key.
• Llama 4 Maverick serving cost: Meta's Llama 4 page gives an estimate of $0.19 per million tokens (3:1 blended) with distributed inference and $0.30 to $0.49 on a single host.

See Model Pricing Explained and How to set spend limits on LLM API keys.

Where are Meta models available?

• Meta Model API: Muse Wonder, Muse Spark and Muse Ember, plus Muse Image, Muse Voice Transcribe and SAM 3.1.
• Meta apps: Muse Spark is available on meta.ai and in the Meta AI app; Muse Wonder will power experiences in the Muse app, rolling out first to subscribers, per Meta's announcement.
• Muse Code: Muse Wonder is available to developers through Muse Code, and Muse Wonder Full to Muse Code subscribers.
• Muse Spark Open weights: Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models; API access through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI, per Meta's announcement.
• Muse Glimmer and Llama 4 weights: Hugging Face; Llama 4 also from Meta's Llama downloads page.

See the Self-Hosted LLM Security Guide for running open-weight models.

Restricted-access Meta models

As of 2026-11-15. Source: Meta announcements, models page and Hugging Face model cards.

What is the context window of Meta models?

Muse Wonder and Muse Spark list a 1,048,576-token context window, and Muse Spark Open lists 1M tokens at inference. Muse Glimmer 30B lists 131,072 or more tokens. Llama 4 Scout lists 10M tokens and Llama 4 Maverick 1M. Muse Ember's context window is not stated on the models page checked, but its pricing has a tier at 200K input tokens; Muse Wonder and Muse Spark have no long-context premium. See Context Windows Explained.

Is any Meta model open source?

Meta publishes weights for three current model lines, under different licences:

• Muse Spark Open: Muse Model License, which Meta describes as having limited commercial terms that let many developers use the model for free, and which incorporates a Usage Policy.
• Muse Glimmer 30B: Apache 2.0.
• Llama 4 Scout and Maverick: Llama 4 Community License Agreement, described on the model card as a custom commercial licence.

Muse Wonder, Muse Spark and Muse Ember are API-only. Whether a licence counts as open source depends on its terms, so read each licence before use. See the Open-Weight Model Security Guide and Running LLMs Locally Explained.

Model cards and system cards

As of 2026-11-15.

See Model Cards and System Cards Explained.

Benchmarks

As of 2026-11-15. Provider-reported and independent results are kept in separate columns.

See How to Read LLM Benchmarks.

Deprecations and retirements

As of 2026-11-15. Meta's models and pricing pages list Muse Spark 1.1, 1.2 and 1.3 side by side and state no deprecation or retirement dates for any Muse model. Meta's announcements checked state no end-of-life date for Llama 4.

Without published retirement dates, pin version IDs and watch the models list endpoint for changes. See the AI Model Retirement Guide.

Data handling and security

• Contributor mode: Meta's pricing page states that contributor mode offers discounted prices or more usage in exchange for permission to use the customer's content to train future Meta models; standard mode carries no such statement on that page.
• Default retention: Not stated by Meta on the Model API pages checked.
• Gated downloads: Muse Spark Open and Llama 4 require identity details on Hugging Face, handled under the Meta Privacy Policy.
• Safety work: Meta says Muse Wonder was trained to avoid sharing private information without permission and to decline requests to help with harmful acts, and that it follows its Meta Superintelligence Scaling Framework.
• Rate limits: limits are set per team, not per key, so one leaked key can consume the whole team's quota.

See AI Data Retention and Deletion and Zero Data Retention for AI.

How to choose a Meta model: factual criteria

• Deployment: API-only (Wonder, Spark, Ember) or self-hosted open weights (Spark Open, Glimmer, Llama 4).
• Training use: contributor variants allow Meta to train on submitted content; standard mode does not carry that term.
• Price per token: compare standard, contributor and Ember's two prompt-length tiers.
• Status: Muse Wonder is in preview; Meta recommends Muse Spark 1.3 for new work.
• Licence: Apache 2.0 (Glimmer), Muse Model License (Spark Open) or Llama 4 Community License.
• Context: up to 10M tokens (Llama 4 Scout), 1M on the Muse API models and Spark Open.
• Hardware: Muse Spark Open loads 816B parameters; Muse Glimmer offers quantised builds for smaller hardware.
• Modalities: Glimmer and Llama 4 take text and images; the Muse API models also take video, audio and PDF.

Practitioner checklist

• Pin production code to an explicit Muse version ID and record it in the AI inventory.
• Decide by policy whether contributor variants are allowed, since they permit training on submitted content.
• Track the models list endpoint for new versions and status changes.
• Read the Muse Model License or Llama 4 Community License before deploying open weights commercially.
• Verify downloaded weights come from the official meta-models or meta-llama Hugging Face organisations.
• Plan rate limits per team and alert on unexpected quota use.
• Store Meta Model API keys in a secrets manager and set spend limits. See the LLMjacking Guide.

Standards and references

• Meta: Model API models (accessed 2026-11-15)
• Meta: Model API pricing and rate limits (accessed 2026-11-15)
• Meta: Introducing Muse Spark (accessed 2026-11-15)
• Meta: Muse Spark 1.1 and the Meta Model API (accessed 2026-11-15)
• Meta: Introducing Muse Wonder (accessed 2026-11-15)
• Meta: Muse Spark Open (accessed 2026-11-15)
• Meta on Hugging Face: Muse Spark Open model card (accessed 2026-11-15)
• Meta on Hugging Face: Muse Glimmer 30B model card (accessed 2026-11-15)
• Meta: Llama 4 (accessed 2026-11-15)
• Meta: Llama 4 model card (accessed 2026-11-15)
• Meta on Hugging Face: Llama 4 Scout model card (accessed 2026-11-15)

Related NHI Mgmt Group resources: Reasoning Models Explained, Multimodal LLMs Explained, Self-Hosted LLMs vs Hosted LLM APIs, LLM Security Guide.