Meta
Open-weight and hosted models spanning Llama, Muse Spark Open, and Meta’s paid Muse Wonder and Muse Spark APIs.
Meta advances open multimodal research and releases the Llama open-weight model family used widely in self-hosted AI stacks. Muse Wonder (muse-wonder-1.0, 2026-10-12) is the newer paid Meta Model API peak for longer agentic work, with Muse Code. Muse Spark 1.3 remains the lower-priced closed API. Muse Spark Open (2026-10-15) is a separate weight release: 816B total / 42B active, 1M context, multimodal, under the Muse Model License with limited commercial terms. Combined with PyTorch, Llama still lets organizations train, fine-tune, and serve models without depending on a proprietary API.
Why Meta matters
Meta changed the economics of AI by releasing capable open-weight models, and Muse Wonder extends Meta’s paid API competition with OpenAI and Anthropic above the still-available Spark 1.3 tier. Because Llama models are released as open weights, organizations can fine-tune, self-host, and deploy them in environments where proprietary APIs are unsuitable—on-prem, air-gapped, or cost-sensitive at scale.
Last reviewed: 18 October 2026
Best fit for
When architects typically choose Meta.
- Self-hosting
- On-premises AI
- Fine-tuning
- Cost-sensitive inference
- Open-source stacks
- Paid Meta Model API agents (Muse Wonder / Muse Spark / Muse Code)
Strengths
DataAIHub qualitative scores from 1 to 5, for architects. Not official benchmark results or a public ranking.
- Open weights★★★★★
- Self-hosting★★★★★
- Fine-tuning★★★★★
- Ecosystem (PyTorch)★★★★★
- Hosted API★★★☆☆
Quick facts
- Founded
- 2004
- Headquarters
- Menlo Park, CA, USA
- Ownership
- Public
- Open source
- Yes
- Enterprise
- Yes
- Flagship model
- Llama 4
On DataAIHub
- 4Models
- 6Products
- 5Tools
- 3GitHub
- 2Research
- 4Guides
- 13Benchmarks
Company profile
- Founded
- 2004
- Headquarters
- Menlo Park, CA, USA
- Founders
- Mark Zuckerberg, Eduardo Saverin, Andrew McCollum, Dustin Moskovitz, Chris Hughes
- CEO
- Mark Zuckerberg
- Funding
- Public (NASDAQ: META)
- Ownership
- Public
- Country
- United States
- Primary focus
- Open-weight foundation models, Multimodal research, PyTorch, Meta Model API / Muse Wonder
- Target users
- Researchers, Developers, Enterprises with self-host needs, API builders evaluating Meta Model API
- Revenue model
- Advertising (parent), Open models as ecosystem strategy, Meta Model API usage
- Deployment
- Self-hosted Llama, Muse Glimmer, and Muse Spark Open weights via Hub and serving engines; Muse Wonder and Muse Spark 1.3 via Meta Model API
- Licensing
- Llama Community License; Muse Glimmer Apache-2.0; Muse Spark Open under the Muse Model License with limited commercial terms; Muse Wonder and Muse Spark 1.3 proprietary API
- Open source
- Yes
- Cloud provider
- No
- Website
- https://ai.meta.com
- Confidence
- High
- Source coverage
- 18
Ecosystem
Competes with
- OpenAI
Open Llama stacks are the main self-hosted alternative to GPT APIs.
- Google DeepMind
Competes on multimodal open vs closed model strategies.
- Mistral AI
Fellow open-weight provider competing for European and enterprise self-host.
Works with
- Ollama
Easiest local runtime for Llama models.
- vLLM
High-throughput serving engine for Llama in production.
- LlamaIndex
Popular RAG framework used with Llama backends.
- Muse Spark
Lower-priced closed Meta Model API path under Muse Wonder, alongside open Llama.
- Muse Glimmer
Apache-2.0 ~30B on-device Muse agent weights alongside open Llama.
- Muse Wonder
Newer closed peak on the Meta Model API and in Muse Code, above Muse Spark 1.3.
Recommended for
- Llama models
Pick the right Llama generation for your deployment.
- Fine-tuning
Because Llama models are released as open weights, organizations can fine-tune, self-host, and deploy them where proprietary APIs are unsuitable.
Often paired with
- Hugging Face
Default Hub for distributing and discovering Llama weights.
How Meta evolved
Key moments in chronological order.
- Open source
Meta releases Muse Spark Open weights: 816B total / 42B active, 1M context, natively multimodal, under the Muse Model License with limited commercial terms. This is a separate checkpoint from the closed Muse Spark 1.3 API.
- Model
Meta releases Muse Wonder (muse-wonder-1.0) on the Meta Model API and in Muse Code. It is the newer peak above Muse Spark 1.3. Standard price is $4 / $0.20 cached / $20 per 1M tokens, with no Contributor mode. Preview. Closed weights. Muse Wonder Full is for Muse Code subscribers.
- Model
Agentic and coding upgrade in Muse Code and the Meta Model API. Previously available reasoning modes ship now; max reasoning is withheld pending additional safety testing. This API release did not include weights.
- Open source
Apache-2.0 30B on-device agent model from Meta Superintelligence Labs; distinct from closed Muse Spark and from Llama 4.
- Product
Coding-focused Muse Spark 1.2 and Muse Code terminal agent beta; Meta Model API access expands globally.
- Model
Muse Spark 1.1 + Meta Model API
Paid multimodal agentic model via Meta’s first public Model API preview.
- Model
Next open multimodal generation for self-hosted stacks.
- Model
Llama 3
Major quality jump for open-weight chat and coding models.
- Open source
Llama 2 open release
Openly available weights for commercial and research use.
- Model
LLaMA to researchers
Research weights that sparked the open-model wave.
- Open source
PyTorch released
Becomes the dominant research framework and later the industry standard for training modern neural nets.
Products
- LlamaOpen-weight model family
- Muse WonderNewer paid peak (muse-wonder-1.0, 2026-10-12) via Meta Model API and Muse Code
- Muse SparkLower-priced closed API (Spark 1.3). Muse Spark Open (2026-10-15) is a separate weight release
- Muse CodeTerminal coding agent (beta) with Muse Wonder and Muse Spark 1.3
- Muse GlimmerApache-2.0 ~30B on-device open agent model
- PyTorchDominant deep learning framework
Foundation models
Related tools
Related research
- Meta AI research
Evaluation
- CRAG — Comprehensive RAG Benchmark
Original Paper
GitHub
Related benchmarks
Related rankings
Related guides
- Llama Models
How Llama generations differ for chat, coding, and multimodality.
- Fine-Tuning
Adapt open Llama weights to your domain.
- Large Language Models
Foundations behind open-weight LLM deployment.
- RAG
Self-hosted retrieval stacks built on Llama.
Explore more companies
- Mistral AIFranceMistral AI builds Mistral Large 3 (with Mistral Large 4, a 1.05T-parameter multimodal MoE with 52B active parameters, in public preview since 2026-10-06), Small/Medium commercial tiers, and open-weight models, plus La Plateforme APIs and Le Chat. Mixtral remains an Apache-2.0 self-host option; it is not listed on Mistral serverless API pricing as of 2026-08-16. Regional Endpoints (EU or US) are generally available. Series D (€3B, 2026-09-08) funds sovereign open-weight infrastructure; a 2026-09-10 Cloudera partnership targets hybrid/air-gapped inference and custom models on customer data. Firefox Smart Window (beta, 2026-09-16) uses Mistral models in France and North America with zero data retention.
- Hugging FaceUnited StatesHugging Face provides the largest open platform for discovering, sharing, evaluating, and deploying machine learning models, datasets, and demos. It has become the default collaboration hub for the open-source AI ecosystem—backed by Transformers, Diffusers, PEFT, and managed inference. On 2026-09-03 NVIDIA agreed to acquire Hugging Face for $12.93B; until close (expected H1 2027, pending regulators) Hugging Face remains independent, and NVIDIA says the Hub will stay open to all model builders, clouds, and accelerators.
- AlibabaChinaGlobal technology group whose Qwen team releases competitive multilingual and multimodal foundation models for cloud and open-weight use.
- Sarvam AIIndiaSarvam AI builds open-weight language models trained in India, plus speech, vision, and enterprise products aimed at Indian languages. Sarvam-M (May 2025) is a 24B Apache-2.0 model post-trained from Mistral Small. Sarvam 30B and Sarvam 105B (open-sourced 6 Mar 2026) are reasoning MoE models trained from scratch on IndiaAI compute: 30B powers the Samvaad conversational platform, and 105B powers the Indus assistant. Weights are on Hugging Face and the IndiaAI AI Kosh; the API is OpenAI-compatible.
- AnthropicUnited StatesAnthropic builds the Claude family of foundation models for enterprises that prioritize instruction following, long-context reasoning, coding quality, and predictable, safety-conscious behavior. Constitutional AI and careful system design are core to how Claude is trained and positioned. Claude Opus 5.5 (claude-opus-5-5) shipped 2026-09-22 at $4/$20 per 1M tokens, Sonnet 5.5 on 2026-09-28, and Haiku 5.5 (claude-haiku-5-5) on 2026-10-07 from $0.10/$0.50, completing the 5.5 lineup. As of 2026-09-16, Claude apps merge chat and Cowork into one conversation (Pro/Max first), with Claude Docs, Slides, and Design available from any chat; on 2026-10-08 those three left beta for every plan, including Free, and Claude Dashboards (live dashboards over connected data platforms, paid plans) and Claude Motion (code-built animations, Team and Enterprise) entered beta. Claude Code Projects (beta, 2026-09-17) add a coordinator that delegates to parallel cloud threads. On 2026-09-18 Anthropic announced an embedded-evaluation partnership with Accenture’s Faculty unit.
- DatabricksUnited StatesDatabricks builds the Lakehouse platform and Mosaic AI for training, serving, and governing models on enterprise data—plus open projects like MLflow, Delta Lake, and Unity Catalog. Teams use it when RAG, agents, and fine-tuning need to run next to governed tables. Kimi K3 is available through Unity AI Gateway (2026-08-06) alongside proprietary frontier APIs.