AI at Meta
77 models across 12 families · Latest: Muse Spark (2026-04)
Large-scale open-source AI for social technologies.
AI at Meta's portfolio covers 75 active models across 12 current families, spanning coding, rag, and agents. Open a model detail page to compare provider routes and sourced benchmarks.
Covers 7 workload areas across 75 active tracked models; last verified 2026-06-07.
Use it for
- Teams evaluating coding, rag, and agents across this lab's releases
- Comparing model families before committing to a flagship
- Migration and pricing follow-ups across 75 tracked models
Do not use it for
- Choosing a hosting provider without opening a model page for price ladders
Active models
75
Current models from this lab, excluding deprecated ones
Active families
12
Current model families from this lab
Open catalog
74 open
0 open source / 74 open weights
Lowest output price
$0.030 /1M
Cheapest tracked output across active models, per 1M tokens
Latest dated release
2026-04-08
Muse Spark
Freshness
2026-06-07
Researched 18d ago
Information
Release cadence
Showing 5 recent dated releases (full timeline below). Latest: Muse Spark (2026-04-08).
Where this lab wins
- Coding: 14 tracked models with SWE-bench / HumanEval-style scores.
- RAG: 16 tracked models with ruler / needle retrieval benchmarks.
- Agentic: 7 tracked models with BFCL, tau-bench, and SWE-bench tool-use coverage.
- Long-context: 24 tracked models with context-token or InfiniteBench-class signal.
Flagship quality / price signal
Flagship: Llama 3 8B Instruct (best sourced coding quality-per-dollar in this portfolio).
Coding task grade A · humaneval score 68.2 · cheapest tracked output $0.040 per 1M tokens
AI at Meta is an American AI research organization founded in 2013. Large-scale open-source AI for social technologies. AI at Meta ships 12 model families totaling 77 models, with the most recent release Muse Spark in 2026-04. Notable families include Muse, Llama 3.3, and Llama 3.2. Use it as a stable reference for lab background, release coverage, and follow-up model pages as. View official API endpoints, benchmark performance, and coding/agent fit for every AI at Meta model.
About
AI at Meta develops cutting-edge AI technologies that power social media platforms like Facebook and Instagram. Their work on computer vision, natural language processing, and machine learning enhances user experiences, improves content moderation, and drives innovation. Software engineers and SaaS executives can leverage AI at Meta's open-source tools and research to build more engaging and safe social applications.
Featured models
| Model | Released | Context | Input price ($/1M) | Output price ($/1M) | License | Openness |
|---|---|---|---|---|---|---|
| Muse Spark | 2026-04-08 | - | - | - | Proprietary | Proprietary |
| Llama 3.3 70B | 2025-12-09 | 8k | $0.9 | $0.9 | Llama 3 Community | Open weights |
| Llama 3.2 11B Instruct | 2025-09-01 | 128k | $0.20 | $0.27 | Llama 3 Community | Open weights |
Model families
Recent releases
- Muse Spark- 2026-04-08
- Llama 3.3 70B- 2025-12-09
- Llama 3.2 11B Instruct- 2025-09-01
- Llama 3.2 90B Instruct- 2025-09-01
- Llama 3.3 70B Instruct- 2025-09-01
FAQ
Who founded AI at Meta and when?
AI at Meta was founded in 2013 and is associated with Menlo Park, California, United States.
What models has AI at Meta released?
AI at Meta ships 77 models across 12 families: Muse, Llama 3.3, and Llama 3.2.
Is AI at Meta's technology open source?
Some AI at Meta models are open-weight (Llama 3.3 70B, Llama 3.3 70B Instruct, and Llama 3.3 70B Instruct (free)); others are proprietary (Muse Spark).
Where is AI at Meta headquartered?
AI at Meta is headquartered in Menlo Park, California, United States.
What is AI at Meta known for?
Large-scale open-source AI for social technologies. Its most prominent tracked family is Muse.
How can I access AI at Meta's models?
AI at Meta's models are available via Arcee AI, Alibaba Cloud PAI-EAS, AWS Bedrock, Baseten API, and Bitdeer AI.
Explore related pages
Last reviewed: 2026-06-07. Data sourced from public lab announcements and provider documentation.