Meta AI
Identity card
The company
- Name
- Meta ↗
- Founded
- 2004
- Head office
- Menlo Park, California, United States
- Chief executive
- Mark Zuckerberg
The product
- Flagship model
- Muse Spark 1.1
- Subscription
- free
- API
- 1.25 $ / 4.25 $ input / output, per million tokens
The data
- Law
- United States
Usage
- Web traffic
- not measured
- Developers
- 1.5 %
Key points
Meta AI is built into WhatsApp, Instagram and Messenger, and its new model is among the best. Meta uses conversations to target advertising outside Europe, and its internal rules tolerated "sensual" exchanges with minors before being withdrawn.
- Quick questions in WhatsApp
- Free
- Privacy
- Teenagers
- In-depth work
1.5 % of tokens processed on OpenRouter, the developers' platform, week of 26/10/2026 (7th of 7 assistants present).
Source: Similarweb, data from August 2025 to September 2026, relayed by the press.
Yes for a quick question in the course of a WhatsApp conversation, with nothing to install. Look elsewhere for sustained work or exchanges that must stay private (Claude, Vibe).
Strengths
- 1 billion users through its messaging appsMeta claims more than 1 billion monthly users for Meta AI, driven by its messaging apps and social networks.Source: Meta, results for the 3rd quarter 2025, 29 October 2025 ↗
- A leading modelWith Muse Spark, released in April 2026, Meta ranks 4th of 419 on LMArena and 2nd in the compar:IA arena.Source: LMArena, text ranking, 28 October 2026 ↗ ; compar:IA, French Ministry of Culture, 4 November 2026 ↗
- Improving on safetyThe Future of Life Institute gives it a D+ in summer 2026, 4th among the companies rated, against a D in 2025.Source: Future of Life Institute, AI Safety Index, July 2026 ↗
Weak points
- Conversations used for advertisingSince 16/12/2025, Meta has used conversations with Meta AI to choose advertisements and content, except in the European Union, the United Kingdom and South Korea.Source: The Register, 1 October 2025 ↗
- Internal rules withdrawn after a Reuters investigationReuters revealed in August 2025 rules that allowed "romantic or sensual" conversations with children; Meta withdrew them after publication.Source: Reuters, 14 August 2025 ↗
- Gas for its data centresFor its Hyperion campus in Louisiana, Meta raised its order from the utility Entergy to 10 gas-fired power plants.Source: The Lens, 30 March 2026 ↗
The company Meta
The Meta entry →Company history
Meta is the former Facebook, founded in 2004 by Mark Zuckerberg at Harvard and based in Menlo Park, California. After releasing its Llama models as open access in 2023, the company launched the Meta AI assistant in September 2023 and built it into its apps.
The story continues
In 2025, Mark Zuckerberg reorganises AI research into Meta Superintelligence Labs, entrusted to Alexandr Wang. The first model from this laboratory, Muse Spark, is released in April 2026, this time without open weights.
Sources: Wikipedia, 7 November 2026 ↗ ; Meta, 8 April 2026 ↗ ; VentureBeat, April 2026 ↗
Leadership
Key figures
- Revenue, 3rd quarter 2026
- 64.6 billion dollarsAssociated Press, Meta results, 21 October 2026 ↗
- Planned investment in 2026
- 135 to 145 billion dollarsAssociated Press, Meta results, 21 October 2026 ↗
Meta AI plans
- Meta AIFree0 $
Meta does not publish a price in euros: the amount paid in France depends on the exchange rate and VAT. Pricing pages, up to date as at 9 November 2026.
Content of the 1 plans, prices in France and the United States
VibeMistral Large1.50 $0.50 $
PerplexitySonar2.50 $0.25 $
DeepSeekV4-Pro3.96 $1.32 $- Meta AIMuse Spark 1.14.25 $1.25 $2.8 times the output price of Vibe
QwenQwen3.8-Max6 $2 $
GrokGrok 4.76 $2 $
GeminiGemini 4 Argon10 $2 $
ClaudeOpus 5.520 $4 $
ChatGPTGPT-6 Astra50 $10 $
Copilot: no API of its own. Flagship model of each company, standard rate, up to date as of 9 November 2026.
Prices up to date as of 9 November 2026, read on the companies' pages, or from 2 concordant sources; the date of each record is in the page's price file Methodology.
Meta AI models
- Company
- Meta
- Models offered
- Muse Sparkby default, not specified (official page ↗)Muse Spark 1.3by default, not specified; 25th of 71 for general intelligence (press ↗)Muse Wonderpaid plan, Muse subscribers (personal agent), Muse Code; 15th of 71 for general intelligence (official page ↗)
Links read on 10/11/2026 on the company's pages, press as a fallback. Rank: Epoch AI ECI index (CC BY 4.0). Performance of AI models
Data management
How Meta handles conversations
- Training models on conversationsIn Europe, adults' public posts and comments and exchanges with Meta AI have been used for training since May 2025. Private messages are excluded, unless they are shared with Meta AI.Source: Meta, 14 April 2025, updated on 27 March 2026 ↗
- Opting out of trainingFill in the objection form sent by Meta by notification and by email.Source: Meta, 14 April 2025, updated on 27 March 2026 ↗
- Applicable lawUS law: Meta is a United States company, subject in particular to the Cloud Act, even when the servers are in Europe.Source: Guide d'usage de l'IA pour les agents publics de l'État (DINUM), June 2026 ↗
Score on this criterion: 25 out of 100, joint 7th of 10. Europeans' public posts used for training since May 2025, unless they object; conversations used for advertising outside the EU (12/2025). The ethics ranking
Environment
Environment score
37 8th of 10- Simple query 40 %50
- Advanced queries 20 %17
- Decarbonised electricity 20 %17
- Environmental approach 20 %50
Black line: average of the 10. The environmental ranking
Our view
Meta AI is built into WhatsApp, Instagram and Messenger, and Meta pushes Vibes, a continuous feed of generated videos: the heaviest use of the 10 after Grok. To power its Hyperion campus in Louisiana, Meta has ordered 10 gas-fired power plants. The company publishes a climate report for each data centre and an SBTi-validated target, but nothing per query; Muse Spark announces 10 times less computation than its predecessor, with no energy figure.
Per-query figure published by Meta
Nothing publishedMeta publishes no per-query figure, either for energy or for emissions. 3 of 10 companies publish one: Google, Mistral AI and OpenAI.
Company figures cannot be compared with each other: scopes and methods differ. The "Simple query" criterion of the environmental ranking combines them with external estimates, for lack of a common basis: the methodology sets out the limits.
- Gemini0.24 Wh
Copilotnothing published- Vibe1.14 gCO2e
- Claudenothing published
- ChatGPT0.34 Wh
- Qwennothing published
- Meta AInothing published
- Perplexitynothing published
- DeepSeeknothing published
- Groknothing published
On the Éclairages blog: What is the electricity consumption of AI? ↗ ; Which AIs pollute the most? ↗
Scores and criteria
The ranking →The Celsius score is the average of 3 scores: performance, ethics and environment. Each combines several criteria. Each row places Meta AI among the 10; clicking a criterion opens its definition and the justification for Meta AI's score.
5 public rankings 100 %774th out of 10
Definition. Synthesis of 5 public performance rankings, where models are ranked by tests or by votes from internet users: each ranking gives 100 to the best of the 10 and 0 to the last.
Justification. LMArena: 4th (rank among 419 models); Artificial Analysis: 50 (index v5.0.1); Epoch AI: 161.7 (ECI index); compar:IA: 2nd (rank among 121 models); ARC-AGI-2: 57.5 (% passed).
Content produced 30 %33joint 7th of 10
Definition. Content the assistant produces on request (sexual, misleading images, minors) and results of independent tests.
Justification. Internal rules allowing "sensual" exchanges with children (Reuters, 8/2025), withdrawn; AI characters closed to teenagers (1/2026).
Use of data 30 %25joint 7th of 10
Definition. Use of conversations for training, retention period, storage location, sanctions by the authorities.
Justification. Europeans' public posts used for training since May 2025, unless they object; conversations used for advertising outside the EU (12/2025).
Safety 20 %425th out of 10
Definition. Independent safety evaluations (Future of Life Institute, SaferAI, Phare) and resistance to attacks.
Justification. Grade D+ from the Future of Life Institute (summer 2026, D in 2025), 33% at SaferAI (22% in 2025); Llama 3.1 405B 6th in the Phare ranking (7/2026), but 96% of attacks succeeded at Cisco (1/2025).
Governance 20 %25joint 7th of 10
Definition. Control of the company and deliberate choices: military uses, shareholding, commitments kept or abandoned.
Justification. Refusal of the European code (7/2025); Llama opened to the US military (11/2024); minors' access to AI companions approved by management, according to a complaint (1/2026).
Simple query 40 %50joint 5th out of 10
Definition. Consumption of a typical text query, on the default model served: figure published by the company, third-party estimate, failing that the size of the model.
Justification. Muse Spark announces 10 times less compute than Llama 4 Maverick (April 2026), with no energy figure; estimate: medium level.
Advanced queries 20 %17joint 8th of 10
Definition. Weight of the resource-hungry uses the product promotes: video and image generation, agents, reasoning enabled by default.
Justification. Vibes, an endless feed of generated videos (September 2025); Meta AI appears in WhatsApp, Instagram and Messenger.
Decarbonised electricity 20 %17joint 8th of 10
Definition. Share of low-carbon electricity in the data centres and efforts to increase it: local grid, on-site gas, renewable or nuclear electricity contracts.
Justification. 10 gas-fired power plants ordered for Hyperion in Louisiana (March 2026); Prometheus powered by gas off the grid.
Environmental approach 20 %50joint 4th out of 10
Definition. What the company publishes and commits to: energy and emissions figures, environmental report, targets, model optimisation.
Justification. Climate report for each data centre and SBTi-validated target; in-house MTIA chips; nothing per query.
Compare
or Meta AI or ChatGPTMeta AI is free and built into WhatsApp, Instagram and Messenger; ChatGPT is more complete and better rated on ethics.Celsius score: 48 against 62 or Meta AI or Gemini2 assistants built into consumer apps; Gemini is far better rated on environment.Celsius score: 48 against 71 or Meta AI or VibeMeta AI uses conversations for advertising outside Europe; Vibe keeps data in the Union.Celsius score: 48 against 40Other comparisons
Meta AI or Claude−26Meta AI or Copilot−10Meta AI or Qwen+9Meta AI or Perplexity+12Meta AI or DeepSeek+13Meta AI or Grok+22Celsius score gap.
Ranked 5th of 10 in the assistant ranking, scores as at 11 November 2026. Facts and figures in the entry collected from 7 to 9/11/2026, each with its source; foundation and headquarters: Meta, annual report 10-K 2025 (SEC, 29/1/2026), 9 November 2026 ↗.
Continue in the ranking
The full ranking →CompanyMeta7models measured, of which 4 with an entryIts assistant, its models, its governance and the electricity of its data centres.The entry →ModelMetaMuse Wonder15thout of 71 on general intelligenceThe best ranked for general intelligence among the models Meta AI offers.The entry →Comparison2 AI assistantsMeta AI or ChatGPT48 against 62Celsius score out of 100Meta AI is free and built into WhatsApp, Instagram and Messenger; ChatGPT is more complete and better rated on ethics.The comparison →The hub newsletter
This month's ranking, news on the impact of AI and one figure from the AI Impact Database. One newsletter a month at most, free of charge. Past issues
