Skip to content

Meta

Dossier checked 28 Sep 2026

What it sells now

Newest frontier release

Muse Spark Open

15 Oct 2026 · the lab’s announcement

816B-parameter (42B active) open-weight model on Muse Spark 1.3’s base, under the Muse Model License, with limited commercial terms

Flagship list price, per million tokens

$4.00 in $20.00 out

Muse Wonder (muse-wonder-1.0), Standard mode, released 12 Oct 2026 · pricing page

Muse Spark Open (15 Oct) is newer but sits a tier below, as open weights with no Meta Model API price; Muse Wonder is priced here as Meta’s most capable model.

No earlier price to compare; see the notes below.

Weights

Closed flagship, some open

Partial. Muse Spark is closed. Muse Glimmer, a 30B model released in August 2026, is open under Apache 2.0, and Llama stays under Meta’s community licence, which needs a separate agreement above 700M monthly users. Meta lists a Muse Spark open-weights release on its roadmap, with no date. Source

Where you can buy the flagship

Its other models, not the flagship

Source

8 releases in the year to 30 Oct 2026, 2 the year before · 16 in all since 2023

Our read, not a fact

Split

The question we ask of every lab: is most of its value in the model, which gets cheaper fastest, or in something other work is built on? Our answer: split between the two.

Meta’s strength is three billion users and a strategy of commoditising rivals’ moats with open weights. The 2026 pivot to closed frontier models bets that distribution plus a private model beats giving the weights away.

What would change our readThe closed pivot fails to out-earn the open strategy, or a rival assistant displaces Meta’s apps as the place people spend attention.

01 · What it runs on, and the money

Trains and serves on

It owns stakes in its biggest new buildings and almost none of the chips. NVIDIA agreed a multiyear partnership in February 2026 for millions of Blackwell and Rubin GPUs; AMD agreed up to 6 GW of Instinct MI450 GPUs, the first gigawatt from the second half of 2026; Meta also designs its own MTIA inference chip. It holds 20% of the Hyperion campus in Louisiana, now planned at 5 GW, and 20% of a new 1 GW campus in El Paso; financing partners own the rest. Source

What it depends on Meta depends on NVIDIA and AMD for accelerators and shares ownership of its two largest new campuses with financing partners. Source

Biggest risk Meta is committing $130-145B of 2026 capex, keeps its frontier model closed while promising an undated open-weights release, discloses no AI revenue, and earned less than a year earlier on 28% higher revenue. Source

Revenue
AI revenue is not disclosed separately; Meta reports only Family of Apps and Reality Labs. Group revenue was $60.8B in the quarter to 30 Jun 2026, up 28%, with net income below the prior year at $15.8B. Source
People
75,472 across Meta (30 Jun 2026), down 1% on a year earlier; the total still includes about 8,000 staff affected by the May 2026 reduction. Superintelligence Labs is not broken out Source
Structure
Meta Platforms, Inc., a Delaware public company with dual-class stock. Class B carries ten votes a share; Mark Zuckerberg holds about 61% of the votes against about 14% of the shares (1 Apr 2026). Source

NotesThere is no year-ago price to compare: in September 2025 the Llama API was a free preview with no list price. A Contributor tier at $0.10 in and $0.20 out is cheaper because Meta may train on your data. The often-quoted $600B three-year infrastructure figure is an executive statement; the filed number is the 2026 capex range above.

02 · Releases

data through 30 Oct 2026

  1. Meta 11 Muse Spark Open15 Oct 2026

Each dot is a frontier release at its date. Click one to find it in the full log. Every lab on one chart

The landmarks

  1. 12 Oct 2026 Muse Wonder / Muse Wonder Full First Muse Wonder models: Muse Wonder on the Meta Model API at $4 in and $20 out and in Muse Code; Muse Wonder Full for Muse Code subscribers only APINew series
  2. Jul 2026 Muse Image First image model out of Meta Superintelligence Labs ImageNew modality
  3. 8 Apr 2026 Muse Spark First closed-weight Meta frontier model, free in the Meta AI app and on meta.ai from launch, US first; the API was a private preview ClosedPivot
  4. Apr 2025 Llama 4 (Scout/Maverick) Mixture-of-experts, multimodal OpenMoEMultimodal
  5. Jul 2024 Llama 3.1 First open model at frontier scale Open405B128K ctx
  6. Jul 2023 Llama 2 First commercial open license OpenCommercial
  7. Feb 2023 LLaMA 1 Research weights that leaked and seeded the open scene OpenResearch
The full release logall 16, newest first
  1. 2026
  2. 15 Oct 2026 Muse Spark Open Frontier 816B-parameter (42B active) open-weight model on Muse Spark 1.3’s base, under the Muse Model License, with limited commercial terms Source
  3. 12 Oct 2026 Muse Wonder / Muse Wonder Full Frontier First Muse Wonder models: Muse Wonder on the Meta Model API at $4 in and $20 out and in Muse Code; Muse Wonder Full for Muse Code subscribers only Source
  4. 2 Sep 2026 Muse Spark 1.3 Frontier In Muse Code and the Meta Model API (labelled public preview) from launch Source
  5. 10 Aug 2026 Muse Glimmer 30B open-weight agent model under Apache 2.0 Source
  6. 5 Aug 2026 Muse Spark 1.2 Frontier Opened to all developers through the Meta Model API (labelled public preview) and OpenRouter Source
  7. 9 Jul 2026 Muse Spark 1.1 Frontier Multimodal reasoning for agentic work; 1M context; Meta Model API preview Source
  8. Jul 2026 Muse Image First image model out of Meta Superintelligence Labs
  9. 8 Apr 2026 Muse Spark Frontier First closed-weight Meta frontier model, free in the Meta AI app and on meta.ai from launch, US first; the API was a private preview Source
  10. 2025
  11. Apr 2025 Llama 4 (Scout/Maverick) Frontier Mixture-of-experts, multimodal
  12. 2024
  13. Dec 2024 Llama 3.3 Frontier 70B efficiency update, the tier just below the 405B flagship
  14. Sep 2024 Llama 3.2 First multimodal Llama
  15. Jul 2024 Llama 3.1 Frontier First open model at frontier scale
  16. Apr 2024 Llama 3 Frontier 8B and 70B open models
  17. 2023
  18. Aug 2023 Code Llama Code-specialised variant
  19. Jul 2023 Llama 2 Frontier First commercial open license
  20. Feb 2023 LLaMA 1 Research weights that leaked and seeded the open scene

03 · When to reach for them

practical read

Agent and coding work
Reach for Muse Spark 1.3 for agent and coding work over a million-token context, at $1.25 in and $4.25 out per million tokens (Sep 2026).
Open base to adapt
Pick Muse Glimmer as a base you can adapt and ship under Apache 2.0 (Aug 2026).
On-device
Use Muse Glimmer where the model must run locally: 30B parameters, on a Mac or one consumer GPU (Aug 2026).

04 · Weights, across all ten

the flagship

Open flagship
DeepSeek Mistral AI Alibaba (Qwen) Moonshot AI Zhipu (Z.ai)
Closed flagship, some open
OpenAI Google Meta xAI
Closed
Anthropic

05 · Essays that read Meta

from the archive

The Forest Floor Is The Product Most people are optimising for canopy. The work that survives the next five AI releases is built from the forest floor up. The Model Is Not the Moat If frontier capability keeps centralising, the durable edge shifts outward into trust, workflow fit, and the surrounding package. The Five Laws of Durable Systems What still has a job after the change? Five tests for seeing what is likely to survive.

06 · About Meta

at a glance

Parent
Meta Platforms (2004)
HQ
Menlo Park, USA
Leadership
Mark Zuckerberg
Division
Superintelligence Labs
Also builds
Facebook, Instagram, WhatsApp
Known for
Llama open-weight models

Meta Platforms (founded as Facebook, 2004, Menlo Park) builds AI through Meta Superintelligence Labs. It became known for the open-weight Llama family, the most widely adopted open LLMs, before moving its 2026 frontier models to closed weights. In August 2026 it released open weights again below the frontier (Muse Glimmer, Apache 2.0).