About Meta
Meta is an AI and social-technology company headquartered in Menlo Park, California. Their mission is to develop large-model systems, open-weight research and generative-AI platforms in vision, language and agents to advance how people create and interact.
Performance
Recent Gateway performance across models from Meta.
—
No recent data
—
No recent data
—
No recent data
Performance metrics will appear after this lab’s models serve Gateway traffic.
Latest models
Recent releases with availability, context, pricing, and usage details.
Limited-time free Together offer for Muse Spark Open. Ends October 18, 2026 at 19:00 UTC. Requires an eligible Together account.
Muse Spark Open is Meta’s open-weight multimodal reasoning model for coding and long-horizon agentic work. It uses 816B total parameters with 42B active per token, supports a 1M-token context window, and includes additional safety post-training for standalone deployment.
Meta’s higher-capability Muse Wonder variant for the hardest coding and long-running agentic work, running at maximum reasoning. Available only through Muse Code with Power Usage or Maximum Usage subscriptions.
Muse Wonder is Meta's agentic model for coding and multi-step work, with native tool calling and MCP support. It plans long tasks, delegates work, and recovers from errors.
Muse Spark 1.3 is Meta's agentic model for long-horizon coding and multi-step work. It improves tool, browser, and computer use across agent harnesses, adds native tool calling and MCP support, and uses fewer tokens than Muse Spark 1.2.
Muse Spark 1.3 Contributor has the same model and capabilities as Muse Spark 1.3 at lower prices. Inputs and outputs may be used to train and improve Meta's AI models.
Muse Voice Transcribe 1.0 is Meta's real-time audio perception model for multilingual speech recognition, speaker diarization, and endpointing. It supports seamless code-switching, keyword and language biasing, audio longer than one hour when streamed, and more than 20 speakers.
Muse Glimmer 30B is an open-weight, dense multimodal model from Meta Superintelligence Labs, distilled from Muse Spark for local agentic workflows. It accepts interleaved text and images and produces text, with support for long-horizon reasoning, tool use, coding, failure recovery, and more than 100 languages.
Muse Spark 1.2 Contributor is the lower-cost, training-enabled edition of Meta's coding-optimized model for agentic workflows. It shares the capabilities and 1 million token context window of Muse Spark 1.2, while allowing Meta to use prompts and completions to train future models.
Muse Spark 1.2 is a coding-optimized model from Meta Superintelligence Labs built for agentic workflows, with improvements to code generation, debugging, and codebase understanding. Its 1 million token context window is designed to handle an entire project in one session.
Muse Spark 1.1 is a multimodal reasoning model from Meta Superintelligence Labs built for agentic tasks, with improvements in tool and computer use, coding, long-context work, and multimodal understanding. It supports a 1 million token context window and is available through the Meta Model API public preview.
Muse Image is Meta's agentic image generation and editing model for precise edits, multi-reference composition, and complex visual prompts.
Around the web
Official profiles, research, and community channels.
